Development Of An Accurate Measurement Algorithm And Device For Fish Morphological Features Based On Deep Learning

In aquaculture,the morphological characteristics of fish such as body length,body width,eye and pupil diameter and tail handle length and width are important indicators for evaluating fish body quality in intelligent aquaculture.Through the measurement of these characteristic indicators,we can know the growth status of fish,which can be used as an important reference basis for classification,feeding,fishing and genetic breeding,and plays an important role in the management and decision-making of fish culture.Therefore,the judgment of fish growth is inseparable from the accurate measurement of fish morphological characteristics.The result obtained by traditional manual measurement method has large error,time-consuming and laborious.In view of the above problems,this paper applies the deep learning method to the intelligent breeding of fish,proposes an accurate measurement scheme of fish morphological features based on deep learning,and designs and implements an automatic measurement device of fish morphological features.The main research contents of this paper are as follows:(1)The data set of fish morphological characteristics was constructed.An efficient and accurate deep learning model is inseparable from a high-quality data set.As the input of the deep learning model,the quality of the data set is closely related to the quality of the trained model.This paper constructs the data set,expands and enhances the data set,so as to increase the number of data sets,simulate different environments in practical application through different transformations,and effectively improve the generalization ability and robustness of the model.(2)A fish morphological feature segmentation and measurement scheme based on solov2 is proposed.Solov2 algorithm is superior in many example segmentation algorithms.By using the fish morphological feature segmentation model required by solov2 training,and inputting the fish image used for testing to the trained model for segmentation and measurement experiments,better segmentation and measurement results are obtained,which solves the problems of large manual measurement error and low efficiency.Then,in order to more accurately measure the fish body length and width,the segmentation effect is significantly improved by testing the expansion and enhancement of the data set and the improved scheme of solov2.After the measured data are counted,it is found that the relative error of body length and width has been further reduced.(3)Design and implementation of automatic measuring device for fish morphological characteristics.Through the design of automatic measurement device,combined with the trained deep learning model,the time of feature measurement is greatly shortened,the measurement efficiency is improved,and the measurement result error is smaller and more accurate.It solves the problems of low efficiency and long time consumption of traditional manual measurement.Therefore,by adopting the automatic measurement scheme of fish somatic characteristics based on deep learning,the problems of large error and low efficiency caused by manual measurement in traditional aquaculture are solved to a great extent,which is of great significance to the scientific and intelligent aquaculture of fish.