Research Of Fish Age Estimation Using Statistical Learning

Fish's age knowledge is an important indicator of fish stock appraisal and management.The traditional way is that fish scientists determine the age of fish according to otolith calcification in the formation of the structure.However,otolith's formation,growth affected by many factors,for example:Season change, temperature, habitat and diet etc., Otolith structure has its own characteristics,only well experienced fish expert can estimate fish's age through otolith,Even so,will have a problem of inaccurate estimates of the age.Therefore,the artificial cognition fish's age remains a time-consuming, laborious problem.Many factors determine that the age of the fish recognition of the urgent need for automation.The use of image processing and pattern recognition, to develop computer-aided automatic identification system of fish age have great value.Based on the otlith image to estimate fish's age is the present fish age research hot spot.an effective feature selection and classification on the otolith image play an important role on the system of fish's age recognition.In this paper,based on the characteristics of the otolith image,we proposed principal component analysis(PCA)and kernel principal component analysis(KPCA)to select the feature of ololith.combined with the classification principle and the decision-making methods of pattern recognition,we proposed support vector machines(SVM)and probability output support vector machines(PPSVM)to classify the extracting features of the otolith image.Including:1. Fish age recognition based on PCA and SVM.This method first obtains the features of fish's otolith image,then through the PCA to extract the principal component of the features,and then input this principal component to the SVM for training,and then classify fish's age through SVM.2. Fish age recognition based on KPCA and SVM.We use the KPCA to extract the principal component of the features of the otolith image.The experiment proved that KPCA method obtain a better recognition effect.3. Fish age recognition based on KPCA and PPSVM. This method applies probability into SVM, make SVM classification results with the probability characteristic,so that we can not only obtain the class that the sample belongs to,but also obtain the probability that the sample belongs to the class.