Research And Implementation Of Agricultural Pest Detection System Based On Convolutional Neural Network

The development of agriculture is related to the economic and social stability of our country.In agricultural production,pests and diseases have been the traditional problems that plague farmers and agricultural departments,which have a great impact on the yield and quality of crops grown in agriculture.Due to the characteristics of wide distribution and variety of pests and diseases,it is difficult for agriculture to take efficient measures in the prevention and control of pests and diseases.At present,agricultural pest detection generally adopts manual detection mode,and its detection efficiency and accuracy depend entirely on the experience of technicians,resulting in low efficiency and high error rate of traditional agricultural pest detection.In this paper,we design a neural network-based agricultural pest detection model for the current status of agricultural pest detection.By extracting features from the diseased crop images and applying the convolutional neural network model,the types of pests and diseases in the images are identified.This machine learning based pest and disease identification model improves the efficiency and accuracy of pest and disease identification.In this paper,we designed and developed an automated system for automatic identification and detection of agricultural pests and diseases.The system applies Io T technology to dynamically acquire the current status photos of agricultural crops,dynamically transmit and submit the collected images to the server side through 4G/5G network,then detect and analyze these photos at the server side.By extracting some pest and disease image features that appear in crops,after extracting these abnormal image features,machine learning algorithms are applied according to common pest and disease features,and using trained machine models,the types and names of pests and diseases in the pictures are analyzed.At the same time,a data platform of pest detection results in the region is established,and the data statistical analysis of the detected pest types is completed with the regional characteristics of crop distribution as well as the temporal characteristics.The main work of this paper is as follows:(1)Analyzed the current status of agricultural pest and disease detection and proposed the corresponding detection technologyBased on the manual pest inspection method and the equipment-based detection mode,compared with the manual identification mode,the pest identification efficiency and accuracy are improved by applying neural networks to analyze and identify pests and diseases based on the equipment to obtain pest and disease images.According to the current pest detection mode,the functional and non-functional requirements of agricultural pest detection are proposed,and the requirements include monitoring device management,pest data collection,pest analysis,and data statistics.Among them,the monitoring equipment management adopts the geographic information platform to manage the basic information and configuration information of monitoring equipment.Due to the wide distribution of agricultural pest monitoring equipment,the geographic information platform is introduced for equipment management,and the analysis of pest and disease detection points can be understood through the geographic information platform.(2)Design of agricultural pest detection modelThe thesis first analyzes the problems of traditional CNN models in the diagnosis of plant pest images due to the limitation of recognizing plant pest images in complex environments.Based on the Inception model,a pest image diagnosis network model is proposed,including the design of a pest identification model and a pest feature recognition model based on the convolutional neural network.In order to improve model performance and solve the problem of reduced recognition rate caused by environmental factors,depth-separable convolution and cascaded dense modules are introduced into the Inception model to improve model recognition efficiency and accuracy.Finally,model parameter analysis and optimization are conducted to build a high-precision and high-efficiency plant pest convolutional neural network model.Then,the open source dataset Plant Village was selected,and datasets of rice,corn,and grape leaf diseases were collected.By data augmentation and random selection,training sets,validation sets,and test sets were divided to test the diagnosis results of different plant types and different feature diagnoses.The results showed that the accuracy of the improved Inception network model designed in this paper was the highest,reaching 98.97%,which was higher than other common models.Finally,based on the diagnostic model,a plant pest image diagnosis recognition system was designed to achieve fast and accurate recognition and diagnosis of plant pests.(3)Implementation of an agricultural pest detection systemAccording to the system technology selection,the selection of technical framework and programming language used for the development of agricultural pest system is studied.Then the specific implementation process of each module function of the system is described,and the modules are programmed and implemented according to the design of the monitoring equipment management,pest and disease data collection,pest and disease analysis,and data statistics modules of the agricultural pest and disease monitoring system.In each module,the implementation techniques are explained in detail,and the implementation process is illustrated with typical program code.Finally,a comparative validation of the recognition results of the convolutional network is implemented.The accuracy of agricultural pest detection based on convolutional neural networks is demonstrated by comparing the algorithm validation.The performance of the system is also tested,and the test passed to show that the system performance meets the expected goal.In conclusion,the agricultural pest detection system designed by the thesis achieves fully automated processing of pest and disease data image acquisition,transmission,and intelligent analysis.Compared with the manual identification model,this pest and disease identification model based on machine learning algorithm model improves the efficiency and accuracy of pest and disease identification.Agricultural pest and disease detection is based on pest and disease monitoring images,and convolutional neural networks are applied to achieve automatic identification of pest and disease categories.Through the analysis and statistics of pest and disease data,it provides intelligent decision analysis for agricultural pest and disease control.