20 research outputs found
Predict sex in salmonids using motion-triggered cameras and artificial intelligence
As the EU water directive starts to be implemented in Swedish law a new national plan of negotiating new environmental permits for every hydropower plant. This process will most likely result in the building of many new fish passages and ladders around the dams to allow fish and other aquatic life to pass them. Those new passages will have to be evaluated to ensure high effectiveness. To accomplish all those studies a new methodology to census fish in a cost-effective and non-labour-intensive way.
This project aims to develop and test a new model which can predict if a specific salmon or trout is male or female. Further on to compare the new model with already existing census methods used to study migrating species of fish. To collect the data needed for this study a camera unit developed by the company TIVA AB to count fish was placed in a salmon trap in the mouth of UmeÀlven near Obbola, VÀsterbottens lÀn. The pictures displaying salmon and trout from the camera were then annotated in Labelstudio to have a dataset to train the model with. To build the model a pre-built algorithm called Yolov5 was used as a base. This algorithm is an improvement to previous AI-learning algorithms as it only looks at the pictures once which increases working speed in comparison to previous models which looked at every picture multiple times.
The results from the two tests conducted show an accurate model when tested on data from the same camera station where light conditions and other parameters match the training data. When tested on data from another site in Stornorrfors with a different camera setup the results are not as accurate.
Unfortunately, the project suffered from big data losses which made the dataset too small to build a very precise model. However, the results show that it is possible to build a model that can predict the sex of a salmon or trout. This is a step towards identifying unique individuals with the help of AI. When more extensively developed, this method will be a very useful and non-invasive tool to get new insights into the lifecycles of aquatic fauna
Signature Verification Using Siamese Convolutional Neural Networks
This research entails the processes undergone in building a Siamese Neural Network for Signature Verification. This Neural Network which uses two similar base neural networks as its underlying architecture was built, trained and evaluated in this project. The base networks were made up of two similar convolutional neural networks sharing the same weights during training. The architecture commonly known as the Siamese network helped reduce the amount of training data needed for its implementation and thus increased the modelâs efficiency by 13%. The convolutional network was made up of three convolutional layers, three pooling layers and one fully connected layer onto which the final results were passed to the contrastive loss function for comparison. A threshold function determined if the signatures were forged or not. An accuracy of 78% initially achieved led to the tweaking and improvement of the model to achieve a better prediction accuracy of 93%
Automatic Individual Identification of Patterned Solitary Species Based on Unlabeled Video Data
The manual processing and analysis of videos from camera traps is
time-consuming and includes several steps, ranging from the filtering of
falsely triggered footage to identifying and re-identifying individuals. In
this study, we developed a pipeline to automatically analyze videos from camera
traps to identify individuals without requiring manual interaction. This
pipeline applies to animal species with uniquely identifiable fur patterns and
solitary behavior, such as leopards (Panthera pardus). We assumed that the same
individual was seen throughout one triggered video sequence. With this
assumption, multiple images could be assigned to an individual for the initial
database filling without pre-labeling. The pipeline was based on
well-established components from computer vision and deep learning,
particularly convolutional neural networks (CNNs) and scale-invariant feature
transform (SIFT) features. We augmented this basis by implementing additional
components to substitute otherwise required human interactions. Based on the
similarity between frames from the video material, clusters were formed that
represented individuals bypassing the open set problem of the unknown total
population. The pipeline was tested on a dataset of leopard videos collected by
the Pan African Programme: The Cultured Chimpanzee (PanAf) and achieved a
success rate of over 83% for correct matches between previously unknown
individuals. The proposed pipeline can become a valuable tool for future
conservation projects based on camera trap data, reducing the work of manual
analysis for individual identification, when labeled data is unavailable
ATRW: A Benchmark for Amur Tiger Re-identification in the Wild
Monitoring the population and movements of endangered species is an important
task to wildlife conversation. Traditional tagging methods do not scale to
large populations, while applying computer vision methods to camera sensor data
requires re-identification (re-ID) algorithms to obtain accurate counts and
moving trajectory of wildlife. However, existing re-ID methods are largely
targeted at persons and cars, which have limited pose variations and
constrained capture environments. This paper tries to fill the gap by
introducing a novel large-scale dataset, the Amur Tiger Re-identification in
the Wild (ATRW) dataset. ATRW contains over 8,000 video clips from 92 Amur
tigers, with bounding box, pose keypoint, and tiger identity annotations. In
contrast to typical re-ID datasets, the tigers are captured in a diverse set of
unconstrained poses and lighting conditions. We demonstrate with a set of
baseline algorithms that ATRW is a challenging dataset for re-ID. Lastly, we
propose a novel method for tiger re-identification, which introduces precise
pose parts modeling in deep neural networks to handle large pose variation of
tigers, and reaches notable performance improvement over existing re-ID
methods. The dataset is public available at https://cvwc2019.github.io/ .Comment: ACM Multimedia (MM) 202
Analysis of Camera Trap Footage Through Subject Recognition
Motion-sensitive cameras, otherwise known as camera traps, have become increasingly popular amongst ecologists for studying wildlife. These cameras allow scientists to remotely observe animals through an inexpensive and non-invasive approach. Due to the lenient nature of motion cameras, studies involving them often generate excessive amounts of footage with many photographs not containing any animal subjects. Thus, there is a need for a system that is capable of analyzing camera trap footage to determine if a picture holds value for researchers. While research into automated image recognition is well documented, it has had limited applications in the field of ecology. This thesis will investigate previous approaches used for analyzing camera trap footage. Studies involving traditional computer vision and machine learning techniques are reviewed. Furthermore, the datasets and additional feature recognition utilized by the techniques will be explored to showcase the advantages and disadvantages of each process, and to determine if it is possible to improve upon them
Estimating deer density and abundance using spatial markâresight models with camera trap data
Globally, many wild deer populations are actively studied or managed for conservation, hunting, or damage mitigation purposes. These studies require reliable estimates of population state parameters, such as density or abundance, with a level of precision that is fit for purpose. Such estimates can be difficult to attain for many populations that occur in situations that are poorly suited to common survey methods. We evaluated the utility of combining camera trap survey data, in which a small proportion of the sample is individually recognizable using natural markings, with spatial markâresight (SMR) models to estimate deer density in a variety of situations. We surveyed 13 deer populations comprising four deer species (Cervus unicolor, C. timorensis, C. elaphus, Dama dama) at nine widely separated sites, and used Bayesian SMR models to estimate population densities and abundances. Twelve surveys provided sufficient data for analysis and seven produced density estimates with coefficients of variation (CVs) †0.25. Estimated densities ranged from 0.3 to 24.6 deer kmâ2. Camera trap surveys and SMR models provided a powerful and flexible approach for estimating deer densities in populations in which many detections were not individually identifiable, and they should provide useful density estimates under a wide range of conditions that are not amenable to more widely used methods. In the absence of specific local information on deer detectability and movement patterns, we recommend that at least 30 cameras be spaced at 500â1,000 m and set for 90 days. This approach could also be applied to large mammals other than deer