Research On Fish Shoal Density Estimation Method Based On Deep Learning

In intensive aquaculture,accurate and objective acquisition of fish quantity and distribution information can not only provide important reference for feeding,grading and other aquaculture operations,but also provide valuable input for the development of intelligent production management system.However,the traditional methods of fish density estimation,such as machine learning sampling,have some limitations because of their low accuracy and low application range.In view of the above problems,this paper takes Atlantic salmon in far offshore mariculture as the research object,in order to realize the real-time,accurate,objective and lossless density estimation of Atlantic salmon.Combining deep learning with machine vision,a deep hybrid neural network model is proposed.The main work of this paper includes:(1)The influencing factors and characteristics of underwater image are analyzed,which lays a foundation for the image preprocessing.The theory of convolution neural network and dilated convolution neural network in deep learning is studied.This provides a theoretical basis for the proposed fish density estimation method based on the depth hybrid neural network model.(2)In order to solve the problem of color offset and blur of underwater fish image,the color correction strategy and contrast enhancement algorithm are used to enhance underwater fish image.It improves the color offset of underwater fish image,enhances the contrast of underwater fish image,and improves the visual quality of underwater fish image.On the basis of image enhancement,the fish are labeled,and the dataset for underwater fish density estimation is made,and the data set is expanded by data enhancement method.(3)Aiming at the problems of low applicability and low accuracy of traditional machine learning,this paper proposes a deep hybrid neural network model based on multi-column convolution neural networks and a dilated convolution neural network.The real-time,accurate,objective and lossless density estimation of underwater fish is realized.The front-end of the network model uses multi-column convolution neural networks to capture the characteristic information of different size receptive fields.Convolution kernels of different sizes are used in each column to adapt to the angle and shape changes caused by fish movements and the size differences between fish bodies.At the same time,in order to reduce the loss of spatial structure feature information,a wider and deeper dilated convolution neural network is used at the back end.The experimental results show that the accuracy of the proposed model is 95.06%,and the Pearson correlation coefficient between the estimated value and the actual value is 0.99.And compared with the other two common models,its accuracy and other evaluation indexes have been improved,which can provide valuable reference for feeding and other breeding operations.