Study On Automatic Classification Of Fresh Water Fish Images Based On ELM Algorithm

With the rapid development of science and technology,more and more computer technology have applied to all walks of life.There are many kinds of freshwater fish in China,and the traditional manual labor is time consuming and labor intensive,so it is very advantageous to use machine vision technology for aquaculture.In the various research directions of image recognition,fish recognition has always been a rare area.In the previous researches,some studies have addressed fish identification issues,but the practical problems are much more complex.In the fish species research,fish identification is the first step of the study,the primary purpose is to split the fish from the picture area with the complex background.In the existing research,the image preprocessing picture is generally collected in some specific experimental environment,the background is relatively single,and the direction and status of the fish have certain requirements.In this paper,parts of the picture come from the major sites,some from own collections,which expanded the scope of the data samples,to a certain extent,and reduced dependence on fish pictures.The main works of this paper are as follows:(1)A total of 200 images of four common freshwater fish were selected as the study subjects,the sample is randomly divided into 4 sub sets by using the 4 cross validation method,each of which takes a subset as the test sample,and the remaining 3 subsets as training samples.(2)A series of treated images of freshwater fish were obtained by pretreatment of fish's image samples,such as background segmentation,image grayscale,binarization,denoising and contour extraction and so on.(3)In order to fully reflect the characteristics of the fishes,consider from the fish color,size and texture three aspects,a total of 25 eigenvalues are extracted based on the RGB color component,Hu invariant moment,and gray covariance matrix.Finally these eigenvalues were taken dimension reduction processing by the principal component analysis(PCA),and 6principal components were extracted.(4)The particle swarm optimization algorithm is used to optimize the two main parameters of the hidden layer of the limit learning machine,which are the input layer weight and the hidden layer threshold,and the optimized limit learning machine classifier is obtained.The experimental results show that PSO-ELM is better than ELM,SVM and BP inrecognition accuracy and stability,and finally the classification accuracy rate reached 96.67%,which reached the expected target classification,and there is the practical application value to the automatic classification of fresh water fish images.