Research On Grey Wolf Optimization Algorithm Based Lepidoptera Insect Image Recognition

Lepidoptera insects do serious harms to all kinds of forests in our country.It is time-consuming and laborious to identify insect species manually.And then it is difficult to meet the actual forestry production requirements of "early control and less loss".This project aims to study how to use statistical information techniques and computer vision methods to statistically analyze the images for research,obtain the features that reflect the essential characteristics of the insects in the images,and use the appropriate classification algorithm to establish a highly accurate Lepidoptera insect recognition model.Then the model can be used for the automatic recognition of insects.In recent years,computation intelligence has been widely used in the process of model optimization for its outstanding global search potential.Grey wolf optimization algorithm is a good method of computation intelligence,but it also has some shortcomings.In this study,we intend to improve the grey wolf optimization algorithm and then apply it to construct automatic recognition model of Lepidoptera insects.The main conclusions are as follows:(1)Aiming at the shortcomings of grey wolf optimization algorithm that convergence speed is too fast and may fall into local optimum,a level learning mechanism is introduced to balance the exploration and exploitation.At the same time,the method of determining the level number and the updating method of the current optimal individual are improved.The experimental results show that the improved grey wolf optimization algorithm can effectively balance exploration and exploitation.Improved grey wolf optimization is used for feature selection to obtain higher recognition accuracy and less feature dimension.(2)In order to train and test the model,1740 images of 9 species of Lepidoptera insects are collected.At the same time,an open Leeds butterfly dataset is used to carry out experiments in order to study the recognition performance of the recognition model under the small amount of images.Leeds butterfly dataset has 10 categories and 832 images.(3)A traditional texture image recognition model is constructed based on the improved grey wolf optimization algorithm.In this model,a texture feature extraction algorithm based on dominant rotated local binary patterns(DRLBP)is introduced to extract features.An improved grey wolf optimization algorithm is used to select invalid and redundant features.Finally,a probabilistic cooperative representation based classifier(PROCRC)is used for classification.The experimental results show that this model has good recognition accuracy.The recognition accuracy of Lepidoptera insect specimens is 88.82%,while that of Leeds butterfly dataset is 89.3%.(4)An automatic evolutionary convolution neural network recognition model is constructed based on the improved grey wolf optimization algorithm.The improved grey wolf optimization algorithm is combined with convolutional neural network by a certain coding technology.It can automatically evolve the structure and parameters of convolutional neural network.The experimental results show that this model has good recognition accuracy.The recognition accuracy of Lepidoptera insect specimens is 89.74%,while that of Leeds butterfly dataset is 89.08%.