Keywords
Deep Learning; Sustainable Development Goals; Fishing Activity Detection; Overfishing; Trajectory Classification
Start Date
5-7-2022 12:00 PM
End Date
8-7-2022 10:00 AM
Abstract
As marine resources play an important role in different aspects of our life such as economy, food, and in balancing ecosystems, using them in a sustainable way is a must. Recently Machine Learning (ML) methods have been used in many domains, one of the most popular classes of ML algorithms being deep learning. There is a potential to apply these methods also on highly complex and rich fishery data. In this work, we aim to exploit deep learning models to help sustainable use of fish resources and avoid overfishing. The work is done in collaboration with the Norwegian Directorate of Fisheries. We aim to develop ML models to detect fishing activities in real time by classifying trajectories of fishing vessels. Such vessels are providing Automatic Identification Systems (AIS) data, and by combining these data with the mandatory catch diaries of Norwegian vessels, we are in retrospect able to classify the vessels as fishing or not fishing. The vessels’ trajectories are segmented and each segment is labelled. Each segment is a sequence of AIS data, with features such position, speed, course, and other. With a neural network with one layer of LSTM with 128 hidden units, a drop out layer with 0.5 dropout rate and two dense layers with relu and softmax activations we have achieved around 0.9 accuracy on the test set. We will continue with more experiments. The dataset and the code will be available in near future.
Use of Artificial Intelligence for sustainable fisheries
As marine resources play an important role in different aspects of our life such as economy, food, and in balancing ecosystems, using them in a sustainable way is a must. Recently Machine Learning (ML) methods have been used in many domains, one of the most popular classes of ML algorithms being deep learning. There is a potential to apply these methods also on highly complex and rich fishery data. In this work, we aim to exploit deep learning models to help sustainable use of fish resources and avoid overfishing. The work is done in collaboration with the Norwegian Directorate of Fisheries. We aim to develop ML models to detect fishing activities in real time by classifying trajectories of fishing vessels. Such vessels are providing Automatic Identification Systems (AIS) data, and by combining these data with the mandatory catch diaries of Norwegian vessels, we are in retrospect able to classify the vessels as fishing or not fishing. The vessels’ trajectories are segmented and each segment is labelled. Each segment is a sequence of AIS data, with features such position, speed, course, and other. With a neural network with one layer of LSTM with 128 hidden units, a drop out layer with 0.5 dropout rate and two dense layers with relu and softmax activations we have achieved around 0.9 accuracy on the test set. We will continue with more experiments. The dataset and the code will be available in near future.
Stream and Session
false