Design And Implementation Of Jujube Pest Recognition System Based On Android Platform

The jujube pests are complex in variety,rapid in reproduction,short in cycle,and have different hazard symptoms,and the control measures are not the same.The traditional identification method of jujube pests is generally based on the characteristics of the color and texture of the pests when identifying the pests,and then compared with the records in the database to determine what category the jujube pests belong to.Category.Artificial extraction of features can lead to loss of information,so it is crucial to be able to quickly and accurately identify date pests.Therefore,it is crucial to be able to quickly and accurately identify the date pests.With the rise of machine learning and deep learning,deep convolutional neural networks have been widely used in image classification and object detection.An Android-based date pest identification system was developed based on neural network technology.The data set of the date pests used in this paper is built from a common data set of date and pests in the orchard.The training set is increased by shooting the image of the date insect pest.The main work of the thesis is as follows.(1)This paper studies the deep learning target detection algorithm based on the region of interest,and introduces the basic concepts of IoU and evaluation index mAP.The overall framework of the Faster RCNN algorithm and the Faster RCNN algorithm in the deep learning target detection algorithm is described in detail.(2)The system researched the Faster RCNN target detection algorithm.The overall training process of the Faster RCNN algorithm is trained using the stochastic gradient descent method.In the process of training,Faster RCNN algorithm first uses ImageNet pretrained Caffe model for network initialization;then randomly selects samples by setting the ratio of positive and negative sample parameters to train to ensure the proportion of positive samples;finally,by using online difficult sample excavator Introduced into the Faster RCNN algorithm,the Faster RCNN algorithm calculates the loss of the extracted region of interest frame during the training process.Then,positive and negative samples are automatically selected for training according to the calculated loss value,thereby eliminating the parameters of this ratio.(3)Through the comparative analysis of the experimental data on the date insect pests,the results of the Faster RCNN algorithm on the dataset of the date pests after the introduction of the online difficult sample excavator system(OHEM)are verified,indicating that the combination of the two can be to some extent.Increase the mean precision mean mAP.