| The goal of fine-grained image classification is to identify subcategories of different classes,for example,to identify whether a dog in a picture is an Alaskan or a Samoyed,where the key is to find local discriminative regions of visual features.However,fine-grained image classification has proven to be a challenging task in computer vision due to the relatively small differences between fine-grained categories and the large differences within categories.Traditional fine-grained image classification methods require professionals to label local discriminative regions in the image,which is not only inefficient but also difficult to guarantee accuracy.With the development of deep convolutional neural networks,their powerful feature extraction ability is widely used in the field of computer vision.Among them,pyramidal convolution has gained much attention in the field of fine-grained image classification with its multi-scale feature extraction capability.This study the fine-grained image classification problem based on pyramid convolution.The main work and contributions are as follows:(1)A fine-grained image classification model(PSBCNN)based on pyramidal convolution and Sim AM attention(Simple,Parameter-Free Attention Module)is proposed to address the problem that fine-grained image classification methods based on strongly supervised information rely excessively on manual annotation.First,the model uses two feature extractors for the feature representation of the input image,and the feature extractor is chosen as a pyramidal convolutional neural network.Secondly,to tap the local discriminative regions in the image,the Sim AM attention module is added to the pyramidal convolution to model each neuron in the feature,and higher weights are assigned to the neurons with discriminative power.Finally,the extracted features are fused using a bilinear pooling to obtain the final feature representation.The PSBCNN model was evaluated on three fine-grained image benchmark datasets,CUB-200-2011,FGVC-Aircraft,and Stanford Cars,with accuracies of 86.3%,88.3%,and 93.7%,respectively.(2)A fine-grained image classification model(Attention-PCNN)based on hybrid attention mechanism and pyramidal convolution is proposed to address the problems that traditional convolution cannot extract multi-scale features of images and existing fine-grained image classification methods are easily disturbed by background information.First,the model uses a pyramidal convolutional neural network to perform multi-scale feature extraction on the input image.Then,the extracted features are fed into two branches capturing global and local information,respectively,and a hybrid attention mechanism is added to the branch capturing global information in order to reduce the interference of image background information and make the model focus more on the target regions with fine-grained features.Finally,the final loss is obtained by weighting the sum of the losses of the two branches.The Attention-PCNN model was evaluated on three fine-grained image benchmark datasets CUB-200-2011,FGVC-Aircraft,and Stanford Cars,and the accuracy rates reached 87.2%,93.1%,and 94.2%,respectively.(3)Based on the above research,this article designs and implements a fine-grained automatic insect identification system.The system aims to protect the biological security of the country,help customs staff to automatically detect and identify insects in imported wood,and solve the problem of difficulty and low efficiency of manual identification.The system mainly includes core functions such as insect identification,data management,user management,etc.It has been put to the test in the resident customs and has achieved good application results. |
