Analysis And Prediction Of Marine Aquaculture Diseases Based On Machine Learning

China’s mariculture industry is in the leading position in the world,and the prevention and control of mariculture diseases are paid more and more attention.Due to the lack of professional and technical personnel,when the disease occurs,farmers often rely on personal experience or online query to analyze and diagnose the disease,blind medication is easy to delay the disease and expand economic losses.In the process of expanding the scale of aquaculture,many aquaculture institutions and research institutes have collected a large number of disease data of mariculture.These image information data have great significance for the prevention and control of marine breeding diseases.Based on this,we can study the relevant algorithms,train the network model,use the mature image recognition technology to identify marine aquaculture diseases and provide scientific diagnosis and treatment methods,so as to reduce the risk of aquaculture and reduce economic losses.Firstly,the disease features in the marine disease image set are preprocessed.The original data set is processed by the directional gradient histogram and Gabor filter.The disease features of the contour and texture of the disease part of the cultivated shrimp are extracted and added to the training set,The purpose of this operation is to solve the problem of feature loss in model training by adding feature information manually.Then,the network model is trained.Firstly,the data set is expanded and unified size is processed,and the training effect is improved by expanding the training set.Secondly,the lightweight convolutional neural network squeezenet is improved to adapt to the training of shrimp disease data set,and then the model is optimized by network pre training,batch normalization and other methods.Through the above operations,a light-weight neural network model named shrimp squeezenet with high recognition rate is obtained;Finally,in order to further improve the model of shrimp disease identification.Several classical network models are trained by data sets,and then fused by majority voting method.The model after fusion is stable,reliable and has high recognition rate of various diseases.The data set used in the experiment is provided by Qingdao Litu hi tech Information Co.,Ltd.,which includes the data collected by Yellow Sea Water Research Institute of China Academy of Fishery Sciences.Excellent data resources provide a reliable basis for the study of marine aquaculture diseases.Firstly,the original data set is preprocessed,and the processed data set and the original data set are trained by the unified neural network at the same time.The results show that the recognition rate of the model after processing is higher.Secondly,by improving the lightweight neural network model squeezenet,the model with higher recognition rate and less computation is trained.In order to further improve the disease recognition rate of the model.Compared with the first mock exam,the fusion of multiple network models further improves the recognition rate of disease,and has a strong stability.