Research On Algorithms Of Underwater Fish Target Detection And Fish Body Segmentation Based On Deep Learning

My country has a wide variety of freshwater fish and its output is huge,but the fishing rate and deep processing rate are low,which is far behind the developed countries in the world.At present,the domestic focus on pulling net and manual fishing.Due to the lack of underwater target detection technology,the operation efficiency is low and the cost is high;in the field of freshwater fish processing,the current reliance on manual or semi-mechanized operations,and the level of technology is low,the segmentation efficiency and accuracy are not high,the damage to various parts of the fish body is relatively large during the processing,and the compatibility of the segmentation of different types and specifications of freshwater fish is insufficient.With the rapid development of image processing and deep learning technology,the use of artificial intelligence technology to achieve underwater fish target detection and fish body segmentation processing is the key technology to solve the above problems.This thesis aims to realize the rapid detection of underwater fish and the precise cutting of fish bodies,and the research of underwater fish target detection and fish body segmentation algorithms based on deep learning is carried out.The main research contents are as follows:(1)In order to solve the problem of poor accuracy of traditional target detection methods in the detection of dynamic fish targets in complex underwater environments,this paper uses the Deep Learning-based Center Net network for underwater fish target recognition,and the network uses anchor boxes.Covering the detected objects,reducing computational consumption,is conducive to real-time detection and capture;at the same time,using the multi-scale retinal enhancement(MSRCR)algorithm with color restoration to weaken the influence of light on the image of underwater fish targets,improve the image quality,and meet the requirements of underwater low resolution fish detection accuracy and speed requirements.(2)The existing fish data set has sample imbalance,which causes the segmentation model training to be over-fitted,and it is difficult to obtain a large number of complex and diverse images.In order to solve such problems,this paper selects five common freshwater fish images as the research object,extracts the corresponding freshwater fish characteristics,studies the deep learning method suitable for fish image data enhancement strategies,and the fish data enhancement model based on the improved generative adversarial network(GAN)is adopted to realize the reasonable expansion of fish images,and at the same time solve the problem that the traditional data expansion method is difficult to obtain a large number of different images.(3)In order to obtain deep-level freshwater fish pixel information,in the encoding feature extraction stage,the fish body semantic segmentation algorithm can easily increase the complexity of the network structure and increase the amount of calculation;at the same time,the large number of down-sampling and pooling operations of the neural network make some detailed information of the fish body is continuously lost during the convolution process,which causes the performance of the segmentation network to be attenuated,resulting in a decrease in the identification and segmentation accuracy of various parts of the fish body.In order to solve the above problems,this paper uses the Deeplabv3+ algorithm optimized by SEnet to realize the semantic segmentation of each part of the fish body,which reduces the amount of calculation,improves the accuracy of segmentation,and reduces the amount of loss of detailed information.(4)Based on Pycharm,PyQt and Qt Designer,the software for fish detection and fish body segmentation is designed,which realizes the rapid detection of fish targets in the underwater complex environment and the precise segmentation of fish bodies in processing.