| With the country’s vigorous development of the "14th Five-Year Plan" and the rapid deployment of rural revitalization strategies,agricultural intelligence is the only way to the future of agriculture.Blueberries,known as the "king of berries",have been sold in the Chinese market since they were introduced from abroad in the 1980 s,and are deeply loved by consumers.With the advancement of the market and the increase of the scale of enterprise production,the problem of insect pests has intensified,especially the blueberry pests have brought a lot of economic losses to the blueberry market.The problem of agricultural pests can be solved only by detecting pests in time and containing them at the root.The traditional methods are mostly manual screening,which cause not so much resources,but also cannot effectively solve the problem of insect pests.Due to the poor recognizability of blueberry pests,the small features of the pests are not obvious and difficult to be recognized by the human eye,which is a more serious problem.In recent years,researchers have used deep learning technology to solve many difficult problems in the field of target detection.Deep learning technology can mimic the human ability to recognize the characteristics of target objects with high accuracy and real-time performance.This paper uses machine vision technology combined with Open CV to perform image processing on pest images,and constructs a blueberry pest dataset.For the blueberry pest data set in this paper,the Faster R-CNN neural network model is used to train it.At the same time,through comparative experiments,the effects of different feature extraction networks on the recognition performance of convolutional network models are verified,and the model recognition efficiency was improved.The description of the working of the system in this paper is presented below:(1)The technical scheme of blueberry jam pest image acquisition equipment was constructed.The actual production environment is simulated by conveying the samples by conveyor belt,and the shooting equipment and shooting light source are selected by comparison.The blueberry jam pest pictures required for the experiment were taken in the laboratory under natural light,a total of 1,000 original sample pictures,and the original sample pictures were enlarged to 1,750 pictures by image translation,staggered cutting,etc.Processing technology and image segmentation technology to process blueberry jam pest pictures.Label Img tool was used to label the data set,and the blueberry jam pest VOC data set was produced according to the data format of Faster R-CNN.(2)The relevant theoretical knowledge in the field of deep learning is briefly introduced,and the Tensorflow deep learning framework is introduced.The detailed composition of the Faster R-CNN network model and the implementation principle of each composition are explained,and VGGNet is selected as the feature extraction module of the model according to the experimental object.A VGGNet-based Faster R-CNN blueberry jam pest identification model was built,and it was found that the VGG19-based Faster R-CNN had better identification results than VGG16 on the pest dataset,and the accuracy was improved by 0.05 m AP.The average rate of blueberry jam pests The recognition accuracy reached 95%,and the F1-Score reached 91%.Therefore,VGG19 was selected as the front-end network for blueberry jam pest identification research.(3)The blueberry jam pest identification system was designed and developed,and Qt was used to design the system interface,which realized the functions of user login,image collection,and pest identification.Using the test set pictures,the system has been tested for pest identification.The test results show that the system has achieved the task of identifying blueberry jam pests,and has certain functionality and accuracy of pest identification. |
PDF Full Text Request