基于深度学习和稀疏表示的害虫识别算法

Algorithm on Pest Identification Based on Depth Learning and Sparse Representation

对农作物害虫种类和数量进行有效的预测是农作物病虫害防治的关键环节,因此基于捕获害虫的样本图片对农作物害虫进行准确的种类识别可以为病虫害的防治提供一定的先验知识.由于农作物害虫种类和形态的多样性及不同种类害虫颜色和纹理的相似性,增加了害虫分类识别的难度.为提高害虫图像识别的准确率,提出了一种基于深度学习和稀疏表示相融合的方法来进行害虫的检测分类识别,进而实现对农作物病虫害的有效防治.该算法首先利用高效的深度学习caffe框架来构建提取害虫特征的网络模型,然后利用该网络模型来提取训练害虫样本的特征向量,从而建立不同种类害虫的超完备字典,最后采用稀疏表示算法来对测试样本进行分类识别.实验部分对10种常见害虫进行了检测识别,实验结果表明新提出的算法有很好的检测分类效果.

The effective prediction of crop pest species and quantity is a key link in the control of crop diseases and insect pests. Therefore, accurate identification of crop pests based on the sample images of insect pests can provide some prior knowledge for the prevention and control of pests and diseases. Due to the diversity of crop pests and the diversity of species as well as the similarity of color and texture of different species, the difficulty of classification and identification of pests is increased. In recent years, deep learning has become a popular direction in the field of images, whose application in the field of image is also more and more, and the use of depth learning algorithm to extract image features have good results. In order to improve the accuracy of pest image recognition, this paper proposed a method based on the combination of depth learning and sparse representation to identify and identify pests, and then realize effective control of crop diseases and insect pests. To abandon the traditional use of LBP and color texture and other feature extraction method, the algorithm first used the efficient depth learning caffe framework to construct the network model to extract the pest features, and then uged the network model to extract the eigenvectors of the training pests samples, thus establishing the super complete dictionary of different kinds of pests, and finally used the sparse representation algorithm to classify and identify the test samples. The experimental results show that the algorithm proposed in this paper has good detection and classification effect.