Development Of An Experimental Device For Identifying Apple Leaf Disease And Variable Spraying

China is a major producer of apples,with the scale and yield of apple cultivation ranking among the top in the world.The sales price and volume of apples are influenced by the quality of apples.Disease prevention and control in the apple production process plays a crucial role in the quality of apples.There are many types of diseases,with early leaf disease being a typical one.Currently,traditional chemical pesticides are mainly used for disease prevention and control.In response to the problems of uneven application,large pesticide residues,and severe environmental pollution caused by traditional spraying methods,the precision variable spraying technology based on image recognition is used to construct an apple leaf disease recognition model using deep learning technology in machine learning,with early apple leaf disease as the research object.A variable spraying device based on precise identification of apple leaf disease is developed,and the main research content is as follows:(1)Construction of an apple leaf disease recognition model based on Movilenetv3.In response to the significant impact of complex environments on existing disease recognition algorithms,which are mainly based on laboratory research and have few applications in field operations,the study focuses on apple leaf disease images from orchards and standard datasets.Mobilenetv3,VGG16,and RestNet50 algorithms are used to construct an apple leaf disease recognition model.Among them,the Mobilenetv3 algorithm constructs the recognition model with the most stable convergence curve when the learning rate is equal to 0.0001,The standard deviation of the loss curve is 0.0449,the standard deviation of the accuracy curve is 0.0126,the accuracy is 0.9498,and the volume is 12.1M.The model recognition accuracy,model size,the standard deviation of the loss and accuracy curve after convergence are all better than other recognition models.(2)Construction of an apple leaf disease recognition model based on improved Mobilenetv3.In view of the problems such as curve oscillation and overfitting after convergence of the Mobilenetv3 algorithm construction model,this research optimizes the Mobilenetv3 algorithm by improving the attention mechanism,improving the full connection layer,and reducing the operator method,and constructs the apple defoliation recognition model based on the improved Mobilenetv3 algorithm.When the learning rate of the model is 0.0001,the convergence curve is the most stable,the standard deviation of the loss curve is 0.0065,and the standard deviation of the accuracy curve is 0.0037,The model volume is 6.29 M,and the model recognition accuracy,model size,loss,and standard deviation after convergence of the accuracy curve are all better than the unmodified recognition model.(3)Development of a variable spraying experimental device based on the identification model of apple leaf disease.On the basis of improving the construction of Mobilenetv3’s apple leaf disease identification model,a variable spraying experimental device was developed.Mainly responsible for designing the overall structure of the experimental device,designing the control system,designing,verifying,and selecting key components,and completing the trial production of the experimental device;Using Proteus and Matlab software,the control system was subjected to duty cycle and fuzzy control simulation tests,and the control system ran well;The error test of flow control was conducted on the prototype,and the control system error was less than 6%,which had little impact on the spray control performance of the device.(4)Variable spraying experimental device for orchard testing.Collecting images of leaflitter disease in orchards,using Mobilenetv3 and improved Mobilenetv3 models to predict the same disease.The experimental results showed that the improved Mobilenetv3 model had better recognition accuracy and response time than the Mobilenetv3 model;The variable spraying experiment includes 6 sets of variable spraying experiments,with different diseases set separately.The effect of the experiment is identified using image processing methods.The experimental results show that the lowest droplet coverage rate of the device can reach over 50%,and the device can achieve variable spraying according to different diseases,with pesticide control effect exceeding 94.5%.