The Classification Of Butterfly Based On Convolutional Neural Network

As a traditional agricultural country,China produces a large number of crops every year,and the losses of crops due to insects are also huge every year.In life,insect organisms are everywhere,whether for protecting biodiversity,maintaining ecological balance,or ensuring the quality of crops And yield,it is of great significance to identify insects.At present,commonly used insect recognition methods include artificial recognition method,conductivity test method,image recognition method,and near-infrared method.The image recognition method is easy to operate and has a high recognition rate,so it is the main method of insect recognition.Traditional image recognition requires manual complex feature extraction of the target image,which is time-consuming and laborious,and the recognition accuracy is not high,and the generalization ability is poor.With the continuous development of computer hardware,the computer's ability to process image data has been continuously enhanced,and the application of deep learning in image recognition has become more widespread.This paper uses deep learning technology and uses convolutional neural networks to mainly extract and classify butterfly insects.The network model improves its learning ability under continuous iterative training,and the network model is improved to a certain extent to improve recognition.Accuracy.The main research work of this paper is as follows:(1)A total of 207 types of butterfly images of 26,111 natural backgrounds were collected as a classification sample data set,and the data was enhanced to expand the original data set by 10 times.The article selects AlexNet,NIN,VGG16 and other networks for comparative experiments.As an extension and enrichment of the network model,this paper designs a simple nine-layer convolutional neural network to explore the impact of data augmentation on insect recognition systems.(2)An improved convolutional neural network method is proposed: based on the multi-layer feature fusion insect recognition method,combining shallow convolutional layer feature output with deep convolutional feature output,deeper convolutional feature semantic information is more abundant,and shallower The layer convolution feature is more clear in describing the detailed texture of the image and other features,and the feature expression of the network model is made more detailed by means of feature fusion.Compared with the improved four networks and the unimproved network models,the recognition accuracy of each network model is improved.(3)The article implements a dual network framework.Feature extraction is the core of image classification and recognition.A single convolutional neural network is often difficult to learn all the information of the input data.This paper designs a dual convolutional neural network architecture.The features extracted by the two networks are used for feature fusion to obtain more accurate and detailed feature expressions for the input data.The two sub-networks can be the same network or different sub-networks.Through experimental comparison,it is found that the dual-network framework recognizes insect pictures.The accuracy rate is higher than the recognition accuracy rate of a single network,and feature fusion between sub-networks can obtain richer feature information.In this paper,the convolutional neural network is studied in depth,and the convolutional neural network model is improved based on the collected data sets.Through experimental comparison between various network models,a network model more suitable for the current insect butterfly data set is explored.The experiments show that the method proposed in this paper has good practical significance and has certain practical value for the research on feature fusion of convolutional neural networks.