Supporting system for fishery resource management utilizing convolutional neural network

Abstract

In recent years, the amount of parent fish stock of Pacific bluefin tuna is decreasing. For this reason, efforts are underway to protect tuna resources in countries around the world. In Japan, an upper limit value of the catch amount is set, and when there is a possibility that the catch amount may exceed the upper limit value, the fishing operation is refrained to protect tuna resources. The set-net fishing covered in this research is a passive fishing method, it is almost impossible to avoid or catch only certain fish species. In addition, if the fixed net fishing is prohibited, the income of fishermen is lost because of fish species other than tuna aren't also caught. In this paper, we propose a method aiming at compatibility between resource protection of tuna and income guarantee of fishermen. An acoustic image of the 30-minute interval obtained by the fish finder is divided into a plurality of divided images, and whether or not the response of tuna is included by convolution neural network (CNN). Finally, based on the identification result of the divided image, it is judged whether the acoustic data of 30 minutes includes tuna. In this paper, to evaluate the performance of the proposed method, we derived the discrimination accuracy of the proposed method, and we also estimated the tuna resource protection effect when applying the proposed method, and the decrease rate of catch.