Morphological Parameters Detection And Quality Inversion Of Fish Based On Stereovision And Machine Learning

In the breeding and breeding of fish,body weight is a very important performance index and breeding target.Traditional measurement method requires manual weighing after sampling,which is difficult to operate,low efficiency,unreliable results and easy to harm the fish.The body weight of fish is significantly correlated with its morphological parameters,so the body weight can be obtained by measuring the morphological parameters.In this study,the methods of machine learning and stereo vision were introduced to improve the efficiency and accuracy of fish body mass measurement.Firstly,the sparse depth information of fish body was obtained by binocular stereo vision and key point positioning technology based on active shape model,and 3d geometric features of fish body were extracted,and then the morphological parameters were determined in situ.Before measurement,super-resolution reconstruction based on sparse representation is carried out.Reconstruction was performed on the basis of two existing data sets and the data set prepared by this study with the addition of underwater image degradation factors.The results show that the location accuracy of key points is higher and the morphological parameters are more accurate after high resolution reconstruction.Secondly,the method of quantifying the correlation between morphological parameters and body weight was studied.Based on the weighted contribution estimation algorithm of input variables,a hybrid process combining voting method and randomization experiment was proposed,which made the evaluation results of the importance of morphological parameters statistically significant.The similarity between the estimated ranking and the reasonable ranking was judged by the appropriate distance measure function.Finally,the morphological parameters were screened under the guidance of the above analysis results.A robust body mass prediction model was developed by combining Adaptive Chaotic Particle Swarm Optimization(ACPSO)and Extreme Learning Machine(ELM).By comparison,the Adaptive Chaotic Particle Swarm Optimization-Extreme Learning Machine,ACPSO-ELM(ACPSO-ELM)algorithm has good stability and prediction accuracy.To sum up,this study proposed a non-contact measurement process of fish body mass by integrating various technologies,starting from measuring fish morphological parameters,quantifying the correlation between morphological parameters and body mass,and obtaining a robust body mass prediction model.Compared to the previous studies,this research puts forward the quantitative method of shape parameters on the correlation between body mass and creatively to how to analyze the morphological parameters and quality of the body into the problem of how to calculate the contribution of the neural network input variables,and randomized experiment combined with voting algorithm makes the method has the complete statistical significance,The ranking and quantification of morphological parameters are realized.