Shark Research program in Southern California CSU Long Beach loses state funding

This is Malibu artist annual collection of his favorite great white shark moments filmed with his drone over the course of 2023. Watch until the end. It is something nobody has ever seen before. Everything here is presented in 4K and was captured along the coast of Southern, Central, & Northern California.

A renowned shark research center in Long Beach, California is in danger of shutting down after the state pulled funding. The lab does much to dispel fear of the unknown beachgoers feel on sighting a Great White. While not the only shark research center in the state, the lab provides a unique resource to southern california, and lack of the additional state funding will force a major reduction in its activities.

The Shark Lab was established in 1966 when Dr. Donald Nelson joined the faculty at CSULB as part of Marine Biology program. Don's commitment to the study of sensory biology and behavior of sharks was instrumental in furthering our understanding of these animals.

The CSULB Shark Lab, through the direction of Don Nelson produced over 50 scientific publications and trained 21 Masters and 1 Ph.D. student during his 30-year tenure. Many of the research projects conducted by the CSULB Shark Lab ventured to locations like Tahiti, Enewetak Atoll, and Baja, Mexico with funding from the Office of Naval Research and National Geographic Society.

Chris Lowe, who took over after Don Nelson's passing in 1997 as the director of the Shark Lab at Cal State Long Beach, told Fox News Digital that the program monitors great white sharks along Southern California's coastline.

Christopher G. Lowe, PhD received the B.A. degree in marine biology from the Barrington College, Barrington, RI, USA, the M.S. degree in biology from California State University, Long Beach, and the Ph.D. degree in zoology from the University of Hawaii at Mānoa. He is a Full Professor with the Department of Biological Sciences, CSULB, where he is also the Director of the Shark Lab. His research focuses on using and developing new technology to study the physiology and behavioral ecology of sharks.

Some studies shown an overall increase in the great white shark population off the California coast over the last decade. This is attributed to a number of factors, including the Marine Mammal Protection Act of 1972 which helped to restore seal and sea lion populations - a primary food source for great white sharks. Sharks range over the worlds oceans, from Australia to the Arctic, so the local population can flucuate rapidly.

In 2018, the Shark Lab received funding from the State of California which enabled it to significantly expand activities, thanks to efforts by Assembly member Patrick O'Donnell (D-LB), who retired from the Assembly in 2022, to advance shark science and education in California. In addition to a research program, the Shark Lab has created a variety of Beach Safety Education Programs and Outreach Programs. They have created a variety of educational curricula and tools for K-12 classrooms and lifeguards with the goal to provide the best available science information to those who will use it most.

Additionally, the CSULB Shark Lab has designed and participated in numerous outreach programs to communicate with the public, and helped answer questions and provide helpful safety information for the public. Their goal is to provide correct, scientific information for locals and visitors to have the best information to have a safe day at the beach.

Lowe said that after the program launched they have been able to use the funds to tag over 300 juvenile white sharks, with 235 of the sharks with active transmitters.

He said that they chose to tag great white sharks since 97% of bites in California in the last century are from these large predators.

The shark expert explained that the program also has 120 acoustic receivers along the sunny California coastline to "listen" for tagged sharks, as well as drone equipment.

"This enables us to provide lifeguards with data about what sharks are off their beaches. How long are they going to be there, what are they doing, and when are they going to leave," Lowe said.

Lowe said that their research has found that sharks are "around people all the time" and that they "largely ignore people." This was confirmed in a recent study by Stanford University and the Monterey Bay Aquarium which documented that your chance of being bitten by a white shark in California has decreased by 90 percent in the past 50 years.

"We also use drones, and that's been an important part of our monitoring as well, because then we could see where sharks were in proximity to people and then address questions about what the risk is," Lowe said. "And some of our data have indicated that sharks are around people all the time in Southern California. And the sharks largely ignore people."

Lowe’s team will rely heavily on its own drone footage, and will collaborate with fire, police and even some television stations to track sharks. Dr. Ju Cheol Moon, a professor in the computer science department, has developed machine algorithms that can identify surfers, boogie boarders and swimmers, which will streamline the process. See W. Zhang et al., "Deep Learning for Shark Detection Tasks," below.

Seal Beach lifeguards have a partnership with the CSULB team. Lifeguards have recently deployed acoustic receivers at beach area the within the Seal Beach City limits. These receivers record any previously tagged marine animal, including sharks, which come within 500 yards of the receiver (depending on the animal’s transponder).

To collect the data, lifeguards remove the receivers from the water and download the data. They then replace the receivers in the water. The collected data is transmitted to the CSULB Shark Lab for review and is uploaded into its larger database. When there is an indication of a marine animal's presence in Seal Beach's ocean waters, the information will be logged on our marine animal monitoring log.

The shark expert explained that the lifeguards have a wealth of information available to help determine if the beach needs to be shut down.

"If a tagged shark is detected off the beach, lifeguards get that data right away from our real-time buoys, and then they can click on a link that will tell them about where that shark is, how big that shark is, where it's been, and what they'll see is that quite often these sharks have visited other beaches, or they've been at their beach for weeks or months at a time."

He explained that lifeguards no longer need to shut down the beach, they can just post a public warning about a shark's presence.

"In the past, anytime a white shark was seen off a beach, they would close the beach, or they pulled people out of the water," Lowe said. "And now, they don't have to close a beach. They'll post signs warning the public that this is white shark habitat, but because of all the data we've accumulated over the last five years, they don't have to shut the beach down."

Lowe said that the local economy is negatively impacted if a beach is closed.

"Every time they close the Southern California beach, that results in an economic impact on that community," Lowe said. "Just by learning more about the sharks, we've reduced the number of beach closures which have economic impacts on the California community."

"What it's done is it saved our coastal communities a lot of money from unnecessary beach closures," he said.

Lowe said that California has seen a rise in great white sharks in recent years, in part, because of continued conservation efforts.

"Our white shark numbers have been going up steadily, and that's because white sharks have been protected in California since 1994," Lowe said.

He said that an increased number of sharks in California's oceans has "raised a big concern."

"With increasing numbers of sharks, we just didn't know if that was going to increase the probability of people being bitten," Lowe said. "And with young sharks using beaches as their nursery habitat, that raised a big concern because we have a lot of people in the water off Southern California year round and among the sharks."

Lowe said that if they do not raise funding from local supporters, they will be forced to shut down.

He said that funding was cut because of California's "very poor" state budget this year.

"Our program was funded for five years and we received funding in 2018. We stretched that $3.75 million state funding to six years," Lowe said. "We were been very frugal with our funds and getting a lot of good information."

"Our funding runs out in June and because of state budget, is very poor this year," he said. "It doesn't look like we're going to be refunded."

He said that they are seeking private funding from individuals and foundations.

"We use a lot of technology in our shark research, and a lot of those tech companies are based here in California. So we're kind of hoping that maybe some of our big tech companies might be willing to pitch in and help us out," he said.

Resources

California State University, Long Beach (CSULB) Shark Lab
California Department of Fish and Wildlife
Shark Identification
Monterey Bay White Shark facts
National Geographic Kids: Shark Facts
National Oceanic and Atmospheric Administration (NOAA) Shark ID Placard
Heal the Bay developed some frequently asked questions (FAQs) about sharks after a swimmer was bitten in the city of Manhattan Beach. While some of the FAQs are specific to the Manhatten Beach incident, there are others that are relevant to any ocean water.

CG Lowe Publications [researchgate]

W. Zhang et al., "Deep Learning for Shark Detection Tasks," 2021 IEEE Green Energy and Smart Systems Conference (IGESSC), Long Beach, CA, USA, 2021, pp. 1-6, doi: 10.1109/IGESSC53124.2021.9618703.
Abstract: Automatic detection of free-ranging sharks from beach areas is of great importance in maintaining a safe humans-hark interaction. The task is especially challenging due to most existing shark detection methods and the sparsity features of field images collected from Unmanned Aerial Vehicle (UAV). Recently, deep learning has been tremendously successful in various real-world applications such as automatic driving system, object detection, face recognition, medical diagnosis, etc.

In this paper, we propose an automated pipeline of shark detection tasks. In specific, we implement several state-of-the-art object detection models into our shark field data set. These algorithms are Faster R-CNN, Mask R-CNN, Feature Pyramid Network (FPN) and RetinaNet.

We report the quantitative comparison results on the above mentioned object detection models and we also provide some detection example images. The experiments show that the models are capable of making a fast and efficient detection among shark and non-shark objects.
keywords: {Deep learning;Face recognition;Pipelines;Object detection;Feature extraction;Unmanned aerial vehicles;Data models;Object Detection;Shark Recognition;Deep Learning;Convolutional Neural Network},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9618703&isnumber=9618676
I. M. Ali, H. -G. Yeh, Y. Yang, W. Zhang, E. N. Meese and C. G. Lowe, "Improvement of Performance of K-Nearest Neighbors Used to Classify Sharks into Behaviors," 2021 IEEE Green Energy and Smart Systems Conference (IGESSC), Long Beach, CA, USA, 2021, pp. 1-5, doi: 10.1109/IGESSC53124.2021.9618685.
Abstract: One metric used to measure the classification performance of K-Nearest Neighbors (K-NN) is F1-Score. K-NN is used here to classify data into shark behaviors, namely, Resting, Swimming, Feeding, and Non-Directed Motion (NDM). The objective of this paper is to improve the F1-Score of the K-NN. It is proven that applying Ensemble Averaging (EA) based filters on the data, prior to classification, improves Signal Power to Noise Power Ratio (SNR) and sequentially F1-Score. Data resizing and other Signal Processing (SP) techniques are also used to produce more accurate features.
keywords: {Measurement;Conferences;Machine learning;Digital signal processing;Activity recognition;Sports;Signal to noise ratio;Activity Recognition;Shark behavior;K-NN;Digital Signal Processing;Machine Learning;Ensemble Averaging;Unbalanced Data},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9618685&isnumber=9618676
Y. Yang, H. -G. Yeh, W. Zhang, C. J. Lee, E. N. Meese and C. G. Lowe, "Feature Extraction, Selection, and K-Nearest Neighbors Algorithm for Shark Behavior Classification Based on Imbalanced Dataset," in IEEE Sensors Journal, vol. 21, no. 5, pp. 6429-6439, 1 March1, 2021, doi: 10.1109/JSEN.2020.3038660.
Abstract: This paper presents the feature extraction, selection and K-Nearest Neighbors (K-NN) algorithm to classify behaviors of sharks based on the data collected by tri-axial acceleration data loggers (ADLs). Because these behaviors are hard to observe in the wild and do not occur frequently, being able to adequately classify them is extremely challenging. In the proposed scheme, we first employ several transformations to enrich the static and dynamic acceleration data. Then, the enhanced data is converted from time to the frequency domain through the fast Fourier transform (FFT) for noise removal. A modified K-NN approach integrated with feature selection is developed and conducted on the frequency domain data to improve the F1-score for minority classes. The acceleration data of California horn sharks (Heterodontus francisci) gathered through ADLs mounted on the first dorsal fin is used to demonstrate our algorithm. A comparison study shows that the features extracted and selected by our proposed scheme can significantly improve the performance of classification on the imbalanced dataset.
keywords: {Feature extraction;Acceleration;Training;Sports;Machine learning algorithms;Data mining;Biological system modeling;Classification;K-nearest neighbors;feature selection;feature extraction;imbalanced data},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9261357&isnumber=9347831
I. M. Ali et al., "Improvement of Classification of Shark Behaviors using K-Nearest Neighbors," 2020 IEEE Green Energy and Smart Systems Conference (IGESSC), Long Beach, CA, USA, 2020, pp. 1-7, doi: 10.1109/IGESSC50231.2020.9284986.
Abstract: The aim of this paper is to enhance the performance of K-Nearest Neighbors (K-NN) used to classify data collected from an Acceleration Data Logger (ADL) into four shark behaviors, namely, Resting, Swimming, Feeding, and Non-Directed Motion (NDM). It is shown that using Ensemble Averaging (EA) to improve Signal-to-Noise Ratio (SNR), data resizing to reduce unbalanced samples distribution among behaviors, and other signal processing techniques enhance K-NN F1 Scores.
keywords: {Conferences;Signal processing;Acceleration;Signal to noise ratio;Sports;Activity Recognition;Shark behavior;K-NN;Digital Signal Processing;Machine Learning;Ensemble Averaging;Unbalanced Data},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9284986&isnumber=9284967
W. Zhang, A. Martinez, E. N. Meese, C. G. Lowe, Y. Yang and H. -G. H. Yeh, "Deep Convolutional Neural Networks for Shark Behavior Analysis," 2019 IEEE Green Energy and Smart Systems Conference (IGESSC), Long Beach, CA, USA, 2019, pp. 1-6, doi: 10.1109/IGESSC47875.2019.9042394.Abstract: An important step in the study of free-ranging animals is to perform automatic identification and estimation of their natural different behaviors. This task is especially challenging for the species in the aquatic environment, for example, California horn sharks (Heterodontus francisci). Because they are relatively small, demersal, and active in the nighttime. It is quite impossible to conduct the observations in a continuous direct way. The shark lab at California State University Long Beach (CSULB) conducted laboratory trials to quantify acceleration signatures of horn sharks for different behaviors including resting, swimming, feeding, and nondeterministic movement (NDM). Currently, most of the existing methods have applied machine learning algorithms to estimate sharks' different behaviors. However, there is still a lack of an efficient and effective way to conduct automatic prediction. In recent years, deep convolutional neural networks have shown the great promise in various computational biology, bioinformatics and neuroscience areas such as biological image analysis, gene expression pattern representation, 3D neuron reconstruction, automatic tumor detection, etc. In this work, we propose novel deep learning models to automatically classify four different shark behaviors using overall dynamic body acceleration (ODBA) through laboratory trial data sets. In specific, we design three deep convolutional neural networks (CNNs) to make fast and accurate predictions and classification. We perform thorough experiments, and the experimental results demonstrate that our proposed models overall produce the best performance than the prior traditional machine learning methods. keywords: {Activity Recognition;Shark Behavior;Deep Learning;Convolutional Neural Network},URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9042394&isnumber=9042382
S. Karan, E. N. Meese, Y. Yang, H. -G. Yeh, C. G. Lowe and W. Zhang, "Classification of Shark Behaviors using K-Nearest Neighbors," 2019 IEEE Green Energy and Smart Systems Conference (IGESSC), Long Beach, CA, USA, 2019, pp. 1-6, doi: 10.1109/IGESSC47875.2019.9042395.
Abstract: In this paper, we aim to ascertain the behavior of California horn sharks (Heterodontus francisci) based on accelerometer data using a machine learning algorithm called K-Nearest Neighbors. We use digital signal processing techniques such as Fast Fourier Transform (FFT) and Discrete Cosine Transform (DCT) to represent the accelerometer signal in the frequency domain which reduces the data size required for training the classifier, the computation time and the memory resources needed, in addition to improving model accuracy. The shark behavior is classified into four classes namely Resting, Swimming, Feeding and Non-Directed Motion. We compare different combinations of time and frequency domain data on the performance of the algorithm. It is shown that the transform domain data considerably improved the accuracy of the classifier.
keywords: {Activity Recognition;Shark Behavior;KNN;Digital Signal Processing;Machine Learning},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9042395&isnumber=9042382
J. Shi et al., "Acoustic Tag State Estimation with Unsynchronized Hydrophones on AUVs," 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Madrid, Spain, 2018, pp. 1919-1926, doi: 10.1109/IROS.2018.8593589.
Abstract: This paper presents an underwater robotic sensor system for localizing acoustic transmitters when the robot's hydrophones cannot be time-synchronized. The development of the system is motivated by applications where tracking of marine animals that are tagged with an underwater acoustic transmitter is required. The system uses two novel real-time calibration algorithms that improve the accuracy of time of flight (TOF) and time difference of arrival (TDOA) measurements. The first algorithm corrects non-linear clock skews in TOF measurements based on temperature variation. The second algorithm compensates the localized relative clock skew between clocks using a mixed integer linear program. To validate the system's performance, an Autonomous Underwater Vehicle (AUV) was deployed to track a moving tag where GPS data was used as ground truth. Compared to traditional TOF and TDOA filtering methods, the results show that the proposed system can achieve reduction of mean localization errors by 59%, and a reduction of the standard deviation of measurements by 44%.
keywords: {Clocks;Sonar equipment;Acoustics;Temperature measurement;Acoustic measurements;Estimation},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8593589&isnumber=8593358
K. D. Smith, S. -C. Hsiung, C. White, C. G. Lowe and C. M. Clark, "Stochastic modeling and control for tracking the periodic movement of marine animals via AUVs," 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Daejeon, Korea (South), 2016, pp. 3101-3107, doi: 10.1109/IROS.2016.7759480.
Abstract: This paper presents a graph-based model of periodic migrations of tagged fish populations and two multi-AUV stochastic controllers developed to track these fish from the model. The model presented in this paper characterizes patterns in the historical movement of tagged fish and is used to develop stochastic tracking by a “model based control” and a “feedback control”. These two controllers permit swarms of AUVs to track the transition probabilities of the tagged population between vertices of the model. To validate these controllers, a periodic model is developed for a simulated population based on three months of geolocation data from a kelp bass (Paralabrax clathratus), and AUV teams utilizing both controllers are simulated in tracking this population. Results show the viability of stochastic controls for multi-AUV tracking of populations whose behavior is well-approximated by the graph-based model. Preliminary trials with physical AUV systems indicate the plausibility of hardware implementation.
keywords: {Sociology;Statistics;Tracking;Stochastic processes;Control systems;Feedback control;Mathematical model},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7759480&isnumber=7758082
Y. Lin, H. Kastein, T. Peterson, C. White, C. G. Lowe and C. M. Clark, "A multi-AUV state estimator for determining the 3D position of tagged fish," 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems, Chicago, IL, USA, 2014, pp. 3469-3475, doi: 10.1109/IROS.2014.6943046.Abstract: This paper presents a multi-AUV state-estimator that can determine the 3D position of a tagged fish. In addition to angle measurements, the state-estimator also incorporates distance and depth measurements. These additional sensor measurements allow for greater accuracy in the position estimates. A newly developed motion model that better accounts for multiple hypotheses of the motion of a tagged fish is used to increase the robustness of the state-estimator. A series of multi-AUV shark tracks were conducted at Santa Catalina Island, California over the span of four days to demonstrate the ability of the state-estimator to determine the 3D position of a tagged leopard shark. Additional experiments in which the AUVs tracked a tagged boat of known location were conducted to quantify the performance of the presented state-estimator. Experimental results demonstrate a three-fold decrease in mean state-estimation error compared to previous works. keywords: {Boats;Atmospheric measurements;Particle measurements;Tracking;Sea measurements;Sonar equipment;Acoustics},URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6943046&isnumber=6942370
D. Shinzaki et al., "A multi-AUV system for cooperative tracking and following of leopard sharks," 2013 IEEE International Conference on Robotics and Automation, Karlsruhe, Germany, 2013, pp. 4153-4158, doi: 10.1109/ICRA.2013.6631163.Abstract: This paper presents a system of multiple coordinating autonomous underwater vehicles (AUV) that can localize and track a shark tagged with an acoustic transmitter. Each AUV is equipped with a stereo-hydrophone system that provides measurements of the relative bearing to the transmitter, as well as an acoustic modem that allows for inter-AUV communication and hence cooperative shark state estimation and decentralized tracking control. Online state estimation of the shark's state is performed using a Particle Filter in which measurements are shared between AUVs. The decentralized control system enables the AUVs to circumnavigate a dynamic target, (i.e. the estimated shark location). Each AUV circles the target by tracking circles of different radii and at different phase angles with respect to the target so as to obtain simultaneous sensor vantage points and minimize chance of AUV collision. A series of experiments using two AUVs were conducted in Big Fisherman's Cove in Santa Catalina Island, CA and demonstrated the ability to track a tagged leopard shark (Triakis semifasciata). The performance of the tracking was compared to standard manual tracking performed using an directional hydrophone operated by a researcher in a boat. In an additional experiment, the AUVs tracked an acoustic tag attached to the tracking boat to quantify the error of the state estimation of the system. keywords: {Target tracking;Sea measurements;Boats;Acoustics;Atmospheric measurements;Particle measurements},URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6631163&isnumber=6630547
C. Forney, E. Manii, M. Farris, M. A. Moline, C. G. Lowe and C. M. Clark, "Tracking of a tagged leopard shark with an AUV: Sensor calibration and state estimation," 2012 IEEE International Conference on Robotics and Automation, Saint Paul, MN, USA, 2012, pp. 5315-5321, doi: 10.1109/ICRA.2012.6224991.
Abstract: Presented is a method for estimating the 2D planar position, velocity, and orientation states of a tagged shark. The method is designed for implementation on an Autonomous Underwater Vehicle (AUV) equipped with a stereo-hydrophone and receiver system that detects acoustic signals transmitted by a tag. The particular hydrophone system used here provides a measurement of relative bearing angle to the tag, but does not provide the sign (+ or -) of the bearing angle. A Particle Filter was used for fusing these measurements over time to produce a state estimate of the tag location. The Particle Filter combined with an active control system allowed the system to overcome the ambiguity in the sign of the bearing angle. This state estimator was validated by tracking both a stationary tag and moving tag with known positions. These experiments revealed state estimate errors were on par with those obtained by manually driven boat based tracking systems, the current method used for tracking fish and sharks over long distances. Final experiments involved the catching, releasing, and an autonomous AUV tracking of a 1 meter Leopard Shark (Triakis semifasciata) in SeaPlane Lagoon, Los Angeles, California.
keywords: {Receivers;Acoustics;Tracking;Sonar equipment;Particle filters;Robots;Global Positioning System},
URL: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6224991&isnumber=6224548

[1] W. Zhang et al., "Deep Learning for Shark Detection Tasks," 2021 IEEE Green Energy and Smart Systems Conference (IGESSC), Long Beach, CA, USA, 2021, pp. 1-6, doi: 10.1109/IGESSC53124.2021.9618703.

Summary

This work focuses on using deep learning techniques for automatically detecting sharks from images captured by drones flying over beaches. The researchers implemented and evaluated several state-of-the-art object detection models, such as Faster R-CNN, Mask R-CNN, and RetinaNet, on a dataset of drone images containing sharks and other objects like surfers, paddleboarders, and waders.

The motivation behind this research is to maintain a safe interaction between humans and sharks in beach areas. Automatically detecting sharks from drone footage can help monitor their activities and reduce potential risks to beachgoers.

The researchers collected a dataset of over 1,200 high-resolution images from drone surveys conducted along the southern California coast. They treated the problem as a binary classification task, distinguishing between shark and non-shark objects.

After training and fine-tuning the deep learning models on their dataset, they evaluated the performance using metrics like Average Precision (AP) and Average Recall (AR). The RetinaNet model, which employs a novel loss function to address class imbalance, achieved the best overall performance in detecting sharks and non-shark objects.

While the models showed promising results, the researchers acknowledge challenges due to the limited number of training images and the sparse nature of the data, where objects occupy small regions in the images. They plan to explore other object detection models, such as YOLO and SSD, to further improve the performance of automatic shark detection in the future.

AI and CNN techniques Used

The paper discusses the use of three state-of-the-art object detection models: Faster R-CNN, Mask R-CNN, and RetinaNet. Here's a brief description of each model and how it was employed in this research:

Faster R-CNN: Faster R-CNN is a two-stage object detection model. In the first stage, a Region Proposal Network (RPN) is used to generate region proposals that may contain objects. In the second stage, these region proposals are classified and their bounding boxes are refined using a convolutional neural network.

In this research, the authors used pre-trained Faster R-CNN models from the Detectron2 library, which were initially trained on the Microsoft COCO dataset. They then fine-tuned these models on their shark dataset, using different backbones like ResNet-50 and ResNet-101.

Mask R-CNN: Mask R-CNN is an extension of Faster R-CNN, which adds an additional branch for predicting instance segmentation masks for each object. It uses a novel layer called RoIAlign, which preserves spatial information better than the RoIPool layer used in Faster R-CNN.

Similar to Faster R-CNN, the authors fine-tuned pre-trained Mask R-CNN models from Detectron2 on their shark dataset, using different backbone combinations like ResNet with FPN, ResNet with conv4, and ResNet with dilated conv5.

RetinaNet: RetinaNet is a one-stage object detection model that addresses the class imbalance problem during training. It introduces a novel loss function called "focal loss," which down-weights the contribution of easily classified negative locations and highlights the contribution of positive locations.

The authors fine-tuned pre-trained RetinaNet models from Detectron2 on their shark dataset, and they mention that RetinaNet achieved the best overall performance among the three models they evaluated.

In their experiments, the authors used these pre-trained models from Detectron2 as a starting point and fine-tuned them on their shark dataset, which consisted of images captured by drones along the southern California coast. They evaluated the performance of these models using metrics like Average Precision (AP) and Average Recall (AR), considering both shark and non-shark object detection.

The authors note that while RetinaNet performed the best, the effectiveness and efficiency of the models were still limited due to the sparse nature of the images and the limited number of training samples available in their dataset.

Authors

The authors of this work are:

Wenlu Zhang
Xinyi Chen
Dhara Bhadani
Patrick Rex
Yu Yang
Christopher G. Lowe
Hen-Geul Yeh

The authors do not explicitly mention their associations in the paper. However, based on the context, it seems they are affiliated with academic institutions or research organizations working on computer vision and marine biology.

Regarding previous related works, the introduction section mentions a few relevant studies:

Some existing research has combined the use of unmanned aerial vehicles (UAVs) and machine learning techniques for shark detection [6–8]. However, the authors note that these methods did not provide a comprehensive investigation of state-of-the-art object detection models.
Classical object detection methods often involved hand-crafted feature extraction techniques like SIFT [9] and HOG [10], followed by shallow machine learning classifiers like Support Vector Machines [11]. The authors highlight that these traditional models have high computational costs and may not produce robust features.
The authors mention that deep learning has made significant gains in various models, such as Convolutional Neural Networks (CNNs) [12–14], Recurrent Neural Networks (RNNs) [15–18], Transformers [19], and Generative Adversarial Networks (GANs) [20]. They also note the success of deep learning in object detection tasks.

The authors do not provide detailed descriptions of these previous works, but they establish the context and motivation for their study by highlighting the limitations of classical methods and the potential of deep learning approaches for shark detection tasks.

artifacts created for validation or replication of the work

The paper does not explicitly mention any artifacts being created or shared for validation or replication of their work. However, it provides some details about the dataset and experimental setup that could potentially enable reproducibility:

Dataset:
- They collected a dataset of 1241 images with a resolution of 3840x2160 pixels.
- The images were captured by drones flying along the southern California coast between specific locations.
- The dataset contains images with sharks as well as non-shark objects like surfers, paddleboarders, and waders.
- They randomly split the dataset into training (803 images), validation (109 images), and testing (329 images) sets.
Data Preprocessing:
- They performed data augmentation techniques like horizontal/vertical flipping, rotation, hue tuning, and random shifting/rescaling of images.
Implementation Details:
- They used the Detectron2 library from Facebook AI Research, which provides pre-trained models on the Microsoft COCO dataset.
- The models were fine-tuned on their shark dataset using PyTorch 1.1.0 and an NVIDIA RTX 2060 SUPER GPU.
- Specific details like the number of epochs and training time are provided.

While the authors do not explicitly state that they are sharing the dataset or code, the level of detail provided in the paper could potentially allow others to recreate a similar dataset and experimental setup for validation or replication purposes.

However, it's worth noting that the paper does not mention any external links or repositories where artifacts might be shared. Without direct access to the dataset and code used in this work, full replication or validation may be challenging for researchers outside the authors' institutions.

Movements, behavior and habitat preferences of juvenile white sharks Carcharodon carcharias in the eastern Pacific

int-res.com

Kevin C. Weng^1,*, John B. O’Sullivan², Christopher G. Lowe³, Chuck E. Winkler⁴, Heidi Dewar¹, Barbara A. Block¹

¹Tuna Research and Conservation Center, Hopkins Marine Station of Stanford University, 120 Ocean View Boulevard, Pacific Grove, California 93950-3024, USA
²Monterey Bay Aquarium, 886 Cannery Row, Monterey, California 93940-1023, USA
³California State University, Long Beach, Department of Biological Sciences, 1250 North Bellflower Boulevard, Long Beach, California 90840-0004, USA
⁴Southern California Marine Institute, 820 South Seaside Avenue, Terminal Island, California 90731-7330, USA

ABSTRACT: Understanding of juvenile life stages of large pelagic predators such as the white shark Carcharodon carcharias remains limited. We tracked 6 juvenile white sharks (147 to 250 cm total length) in the eastern Pacific using pop-up satellite archival tags for a total of 534 d, demonstrating that the nursery region of white sharks includes waters of southern California, USA, and Baja California, Mexico. Young-of-the-year sharks remained south of Point Conception whereas one 3 yr old shark moved north to Point Reyes. All juvenile white sharks displayed a diel change in behavior, with deeper mean positions during dawn, day and dusk (26 ± 15 m) than during night (6 ± 3 m). Sharks occasionally displayed deeper nocturnal movements during full moon nights. On average, vertical excursions were deeper and cooler for 3 yr olds (226 ± 81 m; 9.2 ± 0.9°C) than young-of-the-year animals (100 ± 59 m; 11.2 ± 1.4°C). Juvenile white sharks are captured as bycatch in both US and Mexican waters, suggesting that management of fishing mortality should be of increased concern.

KEY WORDS: Juvenile white shark · Carcharodon carcharias · Habitat · Diel behavior · Satellite tag · Bycatch

Tanaka KR, Van Houtan KS, Mailander E, Dias BS, Galginaitis C, O'Sullivan J, Lowe CG, Jorgensen SJ. North Pacific warming shifts the juvenile range of a marine apex predator. Scientific Reports. 11: 3373. PMID 33564038 DOI: 10.1038/s41598-021-82424-9

Summary

This study looked at how warming ocean temperatures in the North Pacific have caused young white sharks to shift their typical habitat range farther north. Historically, juvenile white sharks under 2.5 meters in length were mostly found off the coasts of northern Mexico and southern California. However, starting around 2014, these young sharks began showing up in increasing numbers in the Monterey Bay area, much farther north than their usual range.

The researchers analyzed data from electronic tags on juvenile white sharks as well as sea surface temperature records going back to 1982. They found that the preferred temperature range for juvenile white sharks had shifted northward during the 2014-2016 marine heatwave in the Pacific. This allowed the young sharks to expand their habitat into newly warmed waters off central California.

While the increased presence of juvenile white sharks raises some concerns about potential impacts on species like sea otters, the study highlights how climate change is redistributing marine life. Community science projects played a key role in documenting this northward shift, underscoring their value in tracking environmental changes. Overall, the warming ocean has enabled these young apex predators to occupy new regions outside their historical territory.

Authors

The authors represent multiple institutions and have conducted prior related research on white sharks and the impacts of climate change:

Salvador J. Jorgensen is from the Monterey Bay Aquarium and Institute of Marine Sciences at UC Santa Cruz. He has extensively studied white shark movements, behavior and ecology using electronic tagging.
Kyle S. Van Houtan is from the Nicholas School of the Environment at Duke University. He has published research on how climate change impacts marine species distributions.
Eric Mailander is from the Monterey Bay Aquarium and recorded observational data on juvenile white sharks that was used in this study.
Christopher G. Lowe is from California State University Long Beach and has expertise in shark nursery habitat and juvenile white shark ecology in the Northeast Pacific.

Several of the authors like Jorgensen, Van Houtan, and Lowe have previously published work on white shark habitat preferences, migration patterns, and how oceanographic conditions like El Niño influence their movements and distributions, especially for juvenile life stages. The lead author Tanaka has also studied how climate change shifts species ranges.

So this interdisciplinary team combined expertise in white shark biology, biotelemetry data, community science observations, and environmental modeling to examine how warming in the Northeast Pacific has facilitated a recent range expansion of juvenile white sharks into novel habitats off central California.

Search This Blog

The State of California