Research On Rapid Identification And Classification Algtithm Of Rice PlanthoppersS Based On Image Processing

Rice has always been one of the most important food crops,together with wheat and corn,known as the world’s three largest crops.It is an important cereal crop with the highest yield per unit area and the highest total yield in China.More than 30 percent of the country’s grain acreage is planted in rice,and more than 65 percent of the Chinese people rely on it as their main food.The change of rice supply and demand will directly affect people’s life,and the security of rice growth will directly affect the ability of each family to obtain food.However,in each growth stage of rice,it will be affected by many varieties of diseases and pests,resulting in low rice yield and poor harvest.Therefore,the timely control of paddy pests has become an important part of the development of agricultural production.Paddy fields with fewer pests produce more high-quality rice per hectare,when pests are rampant in the rice fields,rice yields are low and quality is poor.It poses a serious threat to the stable development of national agricultural economy.Rice planthopper is one of the most harmful pests in the rice growth process.During the period before and after the outbreak of rice planthopper,rapid and accurate identification and classification of rice planthopper is of great significance for the early prevention and management of rice pests and precise medication.Therefore,this article further screened and summarized the rice pest images,which collected by our team in the experimental rice fields in Pukou District,Lishui District,and Gaochun District in Nanjing,Jiangsu Province in 2016,2018,and 2019.And the image data set was required to carry out related experiments.Research on the rapid identification and classification algorithms of rice planthopper images has laid the foundation for real-time acquisition of rice field insect status information.The research content is mainly divided into the following three parts:(1)Rice planthopper image classification method based on sparse representationIn order to solve the problems of slow image processing speed and low accuracy in traditional image processing methods,a rice planthopper image classification method based on K-Singular Value Decompo-sition and orthogonal matching pursuit sparse representation was proposed.On the basis of the existing image data set,firstly,the image was segmented by threshold value to obtain the single insect image required for the experiment.Then,the single insect images were used as the dictionary atoms to construct the initial overcomplete dictionary.The characteristic signals of the original input image were decomposed by the orthogonal matching pursuit algorithm of sparse representation,and used the K-SVD algorithm to update the complete dictionary.Finally,the insect images were classified by calculating the reconstruction errors of the input images.The classification speed was up to 6 frames per second,and the average classification accuracy reached 93.7%.(2)The image classification method of rice planthopper based on improved SSDIn order to reduce the work of image preprocessing and further improve the automation level of rice planthopper image recognition and classification,a rice planthopper image classification algorithm based on improved Single Shot Multi Box Detector was proposed.Firstly,the "Labelimg" label making software was used to make labels for the images obtained,and the paddy field pests were labeled as rice planthopper and non-rice planthopper,which are used as the data of VOC set.Then,the original SSD network was optimized to remove part of the convolution layer,and only the conv4＿3,conv7 convolution layers and additional convolution layers of the original network were retained.Conv7 was changed into a 3*3 convolution kernel by atrous convolution.Finally,perform model training and image recognition classification under the best image input batch sizes obtained.The classification speed was up to 29 frames per second,and the recognition accuracy reached 92.9%.(3)Classification and recognition of rice planthopper under incomplete insect imageIn order to solve the problem,which caused by incomplete acquisition insect images is incomplete in the process of image acquisition processing,the rice planthopper recognition has a low accuracy and slow speed,an incomplete rice planthopper image classification method based on dictionary learning and SSD was proposed.Firstly,a single rice field insect image was segmented into blocks to obtain a mixed sub-image block set with background information and feature information.Then,the sub-image blocks were used as the dictionary atoms to construct the over-complete dictionary,and initializes and optimizes it.The updated overcomplete dictionary was input into the SSD algorithm as a data set for training,and the best training model was obtained.Finally,the collected images which contained incomplete paddy field insects,were tested on the training set model,and the classification speed could reach 22 frames per second,and the recognition accuracy could reach 89.3%.