"Optimized Bilinear Convolutional Neural Networks for Fine-Grained Image Classification of Power Assets"

"The dataset adopted in this study is constructed as a hybrid collection, which integrates publicly available open-source archives and on-site engineering image data acquired from practical power industry scenarios. The collected data covers three core business segments of the power industry, including power transmission equipment, substation equipment and power generation equipment, covering the full-service scenario of power grid assets. In total, the dataset contains 11,191 raw images and is manually categorized into 81 fine-grained power asset categories with detailed attribute differentiation and accurate label annotation. Figure 3 presents typical representative samples of all major categories in the dataset. Specifically, samples a, b and c correspond to typical power transmission equipment; samples d and e belong to common substation equipment; and samples f, g, h and i illustrate various types of power generation equipment. Such multi-scene and multi-category data composition can effectively verify the generalization ability and practical application value of the proposed model in complex power asset classification tasks."