Research On Rice Pest Identification And Counting Method Based On Deep Learning

Rice is one of the important food crops,and accurate identification of rice pests provides an important basis for rice pest control.Currently,rice pest monitoring technologies in China have evolved from traditional insect radar,field detection lights and manual field surveys to more advanced technologies that can not only detect pests more accurately,but also classify and identify them more precisely.In this paper,we propose a new method for rice pest identification and counting based on image processing technology,and design and develop a rice pest identification system.1.Firstly,a full convolutional neural network based insect image classification algorithm is proposed,which consists of a segmentation phase and a classification phase.The segmentation phase employs a new encoder-decoder network that is integrated into the CRF module for accurate and fine-grained insect boundary detection.The classification phase employs a Dense Net framework that introduces an ECA mechanism for efficient classification of rice insect pests.The results show that the algorithm outperforms other algorithms in all aspects.2.Secondly,a pest image target detection method based on YOLOv5 technology is proposed for the problem of rice pest counting.This paper firstly introduces the principle and characteristics of YOLOv5 technique,and then optimizes and improves it to construct a model applicable to rice pest counting.Experiments were conducted on a self-built rice pest dataset and compared with several other commonly used target detection methods in this paper,and performance analysis was performed in terms of detection accuracy,speed and model size.The experimental results show that the method in this paper has high accuracy and efficiency in rice pest counting task,which is better than other methods.This paper provides a novel and effective technical tool for rice pest control.3.Finally,a rice pest identification system was designed and developed.Using image processing technology,the system can accurately analyze the damage level of the pest situation,identify different kinds of pests,and give corresponding suggestions.The system includes modules of image acquisition,image pre-processing,image segmentation,feature extraction,classification recognition and result display.This paper introduces the system development environment,My SQL database design and system functions and other related contents in detail.In this paper,a new method for rice insect pest detection and identification is proposed,and a rice insect pest identification system is designed and developed.The method and system can effectively solve the shortcomings of existing methods for automatic identification and classification of field insects,such as insufficient data sample size,complicated data pre-processing,insufficient feature extraction,fluctuating model fit and similar target features,and provide a novel and practical technical support for the rice industry.