Research On Recognition Classification And Prediction Of Greenhouse Pests Based On Machine Learning

With the improvement of agricultural technology and production technology and the accelerated adjustment of China’s agricultural production structure,China’s greenhouse industry has maintained steady growth.At present point,the main factors affecting the productivity and quality of crops in the greenhouse are agricultural pests.Accurate pest detection and dynamic pest prediction in greenhouses are essential for encouraging the economy of greenhouses to grow further.Traditional pest management methods generally rely on human observation and arbitrary administration of pesticides,which frequently leads to drug overuse or abuse and lower crop yields.With the rapid development of image processing,computer vision,and machine learning technologies,their application in agricultural pest control has achieved good results.Based on this,this thesis takes greenhouse pests such as whiteflies and thrips as experimental objects,combines and improves image processing and machine learning related technologies and algorithms,studies and analyzes the identification and counting of pests and the dynamic warning of pest outbreaks,and designs an automated identification and warning system.Tests have shown that the system is efficient and accurate in identifying and counting pests,and can effectively predict outbreaks of pests.The main work and innovative points of this thesis are as follows:1.In response to the problem of complex noise in sticky insect board images and difficulty in object segmentation caused by a large number of pests,this thesis designs a complete workflow that comprehensively utilizes techniques such as image enhancement,edge detection,and threshold segmentation to accurately segment the pest target area.Tests have shown that the designed target segmentation process achieves good segmentation accuracy on real sticky board images.Finally,the classification features were extracted from the individual pest images obtained through target segmentation and a training dataset was constructed.Linear support vector machines were used to complete the training of the classification model,and testing showed that the recognition accuracy of the classification model was 96%.2.A pest level prediction model based on BP neural network was constructed.To solve the problem of low efficiency of stochastic gradient descent(SGD)and easy to fall into local maximum,an improved SGD algorithm combining adaptive weight updating and learning rate attenuation was proposed,and the model was evaluated from three aspects: parameter optimization path,loss function and generalization.The experimental results show that the improved algorithm has a certain improvement in model performance.Finally,a comparative analysis was conducted between the proposed model and existing SVM pest prediction models,and the results showed that the proposed model has higher prediction accuracy.3.In order to better apply it to practical projects,this thesis designs and develops an intelligent identification and warning system for greenhouse pests.The system visualizes pest information at different time periods and generates a heat map of pest density;Dynamically predict pest levels based on environmental information,and provide corresponding pest control strategies based on the results.