Motor magnetic shoe defect classification method based on region-of-interest enhancement

Abstract

The invention discloses a motor magnetic shoe defect classification method based on region-of-interest enhancement, and the method comprises the steps: firstly constructing a classification model which is composed of a convolution layer, a feature reduction layer, a maximum pooling layer, four convolution blocks, four conversion blocks with feature reduction, and a prediction layer; then, acquiring a motor magnetic shoe classification training sample set, and training the constructed classification model by using the motor magnetic shoe classification training sample set to obtain a trained classification model; and finally, collecting a motor magnetic shoe surface grey-scale map of a to-be-detected motor magnetic shoe, and sending the motor magnetic shoe surface grey-scale map into the trained classification model, thereby obtaining a category label of the to-be-detected motor magnetic shoe. According to the method, the features of the feature tensor are subjected to recovery and large-range space association, and the region of interest of the motor magnetic shoe defect classification network is enhanced, so that the classification and anti-interference capabilities of the model are improved, and the classification performance and robustness of the classification model are improved.

Description

technical field [0001] The invention relates to the technical field of motor fault detection, in particular to a method for classifying motor magnetic tile defects based on region-of-interest enhancement. Background technique [0002] Convolutional neural networks have achieved great success in image classification. Models such as VGG19, ResNet, and DenseNet all use down-sampling of input images and combine convolutional and pooling layers to output feature maps layer by layer. Compress, extract low-level features and high-level features of images. Although the learning method of expanding the receptive field through the pooling layer can associate the features with a long distance on the input image, and turn the local feature extraction to the global feature extraction, it is easy to cause feature loss in its forward propagation. The main pooling methods used in traditional convolutional neural networks are maximum pooling and mean pooling. The former compresses the featu...

AI Technical Summary

Problems solved by technology

[0004] What the present invention is to solve is that in the existing detection and classification of motor magnetic tile surface defects based on machine vision and deep learning technology, the deep convolutional neural network used is For the problem that tiny features are not sensitive and features are easily lost in the forward propagation, a method for classification of motor magnetic tile defects based on region of interest enhancement is provided

Images