Research On Fish Phenotypic Data Measurement Method Based On Deep Learning

Fish and related products are one of the important sources of protein in human diets.The healthy development of fisheries is closely related to people’s lives.Excellent fish resources are the basis for the sustainable,healthy and rapid development of fisheries.Breeding and optimization are the key links in the fishery breeding process.In addition to relying on some biochemical indicators for fry selection,the phenotypic data of fry is also an important reference index for fry selection.Current fish breeding research institutions mostly use manual measurement methods to obtain phenotypic data of fry.This process is time-consuming,laborious and error-prone,and it is difficult to obtain comprehensive phenotypic data.In order to help researchers quickly and accurately determine the phenotypic data of fish fry and reduce the labor and time consumption in the determination process,this thesis takes channel catfish as the research object,based on deep learning algorithms,to achieve rapid determination of phenotypic data of fish fry in water.Seven phenotypic data should be considered when selecting channel catfish breeding,including body length,full length,head length,body height,tail handle length,tail handle height and body thickness data.In order to be able to quickly and accurately determine the above phenotypic data,this thesis designs a method for measuring fish phenotype data based on deep learning.The main research contents of the thesis are as follows:1.The fish detection algorithm is constructed based on the deep learning method,which can extract the coordinate position of the catfish in the side view image and the top view image in real time.Combined with the improved residual attention mechanism,the feature pyramid structure and the CIo U loss function improve the fish detection performance accuracy.2.Based on the images of catfish in side view image and top view image obtained by the detection algorithm,19 key points of catfish in side view image and top view image are defined.These key points are used to construct the spatial skeleton of catfish.Then,a deep learning model of key point positioning based on the Hourglass Net network is constructed to realize the key point positioning of catfish image,and the fish body mask is used to improve the accuracy of specific key point positioning of the fish body.3.Based on the coordinate values of the key points in the side view image and the top view image,a three-dimensional coordinate space is established,and the real phenotype data of the fish is obtained by using the idea of segmented integral and proportional parameters.Experimental results show that the measurement method proposed in this thesis can achieve an accuracy of less than 4% on the average relative error of body length and full length,and can also achieve other performance data measurements.The measurement time of the whole measurement system for a single catfish is about 1s.The measurement method proposed in this thesis can also meet the phenotypic data determination of other animals,and has good scalability.