基于注意力机制学习的变电设备缺陷检测方法

针对现有的变电站缺陷图像检测识别算法鲁棒性弱问题，提出一种基于注意力机制学习的变电设备缺陷图像检测识别方法。所提方法以卷积神经网络作为缺陷图像特征提取的骨架网络，融合注意力机制原理，进一步提升缺陷图像特征的可辨识性。首先，构建注意力机制的卷积神经网络特征提取模型，提取不同注意力机制下变电站缺陷图像特征；其次，设计一种自适应特征学习函数，将不同注意力机制下的特征融合成为新的高质量变电缺陷图像特征；最后，将不同注意力机制下的缺陷图像特征输入到分类模型，实现变电站缺陷图像检测。所提方法增强了变电设备缺陷图像检测的准确性与鲁棒性，实验结果显示，所提方法的mAP达到了70.4%。

Method of Substation Equipment Defect Detection Based on Attention Mechanism Learning

In order to solve the problem that the existing algorithm of substation defect image detection and recognition cant work effectively, a method of defect image detection and recognition of substation equipment based on attention mechanism learning is proposed. The proposed method uses convolutional neural network as the skeleton network of defect image feature extraction, and integrates the principle of attention mechanism to further improve the recognition ability of defect image features. Firstly, the convolution neural network feature extraction model of attention mechanism is constructed to extract the features of substation defect image under different attention mechanisms; secondly, an adaptive feature learning function is designed to fuse the features into new high-quality substation defect image features under different attention mechanisms; finally, the defect image features under different attention mechanisms are input into the classification model to realize the detection and recognition of substation defect image. The proposed method can improve the accuracy and robustness of defect detection. Extensive experiments show that the accuracy mAP of this method is 70.4%.