针对变电站设备巡检图像识别领域中存在的图像规模有限和识别准确率低等问题，提出了一种基于改进ResNet34网络的图像分类识别方法。采用Seam Carving算法对图像中的低能量区域进行压缩以保留关键特征；同时使用弹性变换、高斯噪声等6种图像增强技术来增强图像的多样性。将基础ResNet34网络与卷积注意力模块结合，增强模型对设备巡检图像关键特征的提取能力。使用在ImageNet数据集上的预训练模型作为迁移学习的特征提取器来解决样本数量不足的问题。在Adam优化器中引入余弦退火策略来动态调整学习率使改进的ResNet34网络更快收敛至最优解。试验结果表明所提方法比基础ResNet34网络的准确率提升了0.073 3，损失率降低了0.201 9，为变电站设备巡检图像识别领域提供了一种可靠的解决方案。

Aiming at the problems of limited image scale and low recognition accuracy in the field of substation equipment inspection image recognition, an image classification and recognition method based on improved ResNet34 network is proposed. The Seam Carving algorithm is employed to compress the lowenergy areas in the image for the preservation of key features. Additionally, six types of image enhancement techniques such as elastic transformation and Gaussian noise are utilized to increase the diversity of the images. The basic ResNet34 network is integrated with the convolutional block attention module to enhance the model′s ability to extract key features from equipment inspection images. A model pre-trained on the ImageNet dataset is utilized as a feature extractor for transfer learning to address the issue of insufficient sample quantity. A cosine annealing strategy is introduced in the Adam optimizer to dynamically adjust the learning rate, to make the improved ResNet34 network converge to the optimal solution faster. Experimental results show that the proposed method improves accuracy by 0.073 3 and reduces the loss rate by 0.201 9 compared to the basic ResNet34 network, which provides a reliable solution for the field of substation equipment inspection image recognition.

云南省重点研发计划(202303AA080002)；云南省基础研究重点项目(202301AS070055)；云南省基础研究青年项目(202201AU070086)