Fish abundance estimation from multi-beam sonar by improved MCNN

Aquatic products provide essential food and nutrients to humans, the abundance of fish is used in many aspects of aquaculture management, and it also undertakes a lot of tasks in the process of aquaculture, so it is a crucial link. This study introduces an automated method for estimating fish abundance in sonar images based on the modified MCNN (multi-column convolutional neural network), named FS-MCNN. We also proposed the multi-dilation rate fusion loss, which will improve the accuracy and robustness of the model. This method will improve the impact of low pixels in sonar images and blurry edges of target objects in sonar images. It further decreases the RMSE by 14.22% and the MAE by 11.83%, and the final accuracy is 92.83%. This study estimates fish abundance through imaging sonar, which will be able to reduce the effect of light and the complex environment, and it will also contribute to increase labor productivity, reduce the feed waste and enhance the level of information technology in aquaculture or fisheries.