Research On Recognition Method Of Pleurotus Ostreatus Stick Contamination Based On Deep Learning

With the support of the country for agricultural and rural development,the development of the edible mushroom industry has made unprecedented breakthroughs,and the mushroom planting industry has rapidly developed.In response to the problems of untimely selection and treatment of contaminated pleurotus ostreatus stick during the cultivation and growth process of pleurotus ostreatus stick,the main reliance on manual selection of contaminated sticks,and the lack of information management systems in enterprises,a pollution identification model for pleurotus ostreatus stick was established based on the key technology research and industrialization of intelligent factory production of edible mushrooms,a major scientific and technological innovation project in Shandong Province,using deep learning technology,software development,and other technologies,We have designed and implemented a pleurotus ostreatus stick contamination identification system,breaking through the key technology of intelligent identification of stick contamination,and ultimately achieving accurate identification and judgment of stick contamination status.The main research content and results are as follows:(1)A dataset of pleurotus ostreatus stick contamination was constructed and data augmentation was performed.Through on-site research,image data of pleurotus ostreatus stick was collected in a smart factory cultivation shed of a company in Shandong.The distance between the camera and the sticks was controlled at about 60 centimeters,and four photos were taken for each stick.Finally,942 images of Aspergillus flavus contaminated sticks,893 images of Trichoderma contaminated sticks,804 images of Mucor contaminated sticks,and 1118 images of normal sticks were collected,totaling 3757 images.In order to prevent the problem of overfitting of model training caused by the unbalanced number of samples,the data set was enhanced.Finally,the expanded data was divided into training set and test set according to 8:2.(2)An improved ResNeXt50 pleurotus ostreatus stick contamination identification model was constructed.Firstly,experiments were conducted on four common image classification models on a self built dataset of pleurotus ostreatus stick contamination.The experimental results showed that VGG16 had the lowest accuracy of 85.31% on the training set,while ResNeXt50 had the highest accuracy of 90.31%.In addition,the ResNeXt50 model also had high efficiency.Therefore,the ResNeXt50 model was selected for further optimization to improve the recognition accuracy of the model for pleurotus ostreatus stick contamination.When improving ResNeXt50,taking into account the impact of attention mechanism on model accuracy,experiments were designed to explore the impact of ECANet’s embedding method on the accuracy of pleurotus ostreatus stick contamination recognition.Three methods were studied,namely serial embedding,parallel embedding,and residual embedding.The experimental results showed that the parallel embedding method had the most significant impact on pleurotus ostreatus stick contamination recognition,with an accuracy improvement of 8.11% compared to before the improvement.(3)The design and implementation of a contamination identification system for mushroom sticks have been completed.Based on the improved ResNeXt50 model,this paper designs and develops a pleurotus ostreatus stick contamination identification system based on the front-end Vue framework and the back-end Spring Boot framework.Completed the requirements analysis of the pleurotus ostreatus stick pollution identification system,the overall framework and functional module design of the system,the E-R design of the system database,and the physical design of the database.The system includes main functional modules such as basic information setting,pleurotus ostreatus stick management,pleurotus ostreatus stick pollution identification,and pollution stick feedback statistics,achieving automatic identification of contaminated pleurotus ostreatus stick Collecting,inputting,storing,and managing the collected pollution image data,and conducting statistical analysis of stick contamination information have solved the long-standing problem of inefficient manual stick inspection,and helped enterprises develop towards agricultural informatization.