Abstract: To address the low detection accuracy of infrared image detection algorithms for substation equipment caused by equipment occlusion, multiscale objects, and complex backgrounds, this study proposes a substation equipment infrared image detection model based on an improved Faster R-CNN. Initially, the feature extraction network was enhanced based on VGG16 by streamlining certain high-level convolutions, thereby accelerating the training and testing speeds of the model. In addition, the CloBlock module was introduced during the feature extraction phase to further enhance the model's feature-learning capabilities. Subsequently, anchor boxes with aspect ratios of 1:3 and 3:1 were added to improve the detection accuracy of elongated devices. Finally, multiscale region of interest (RoI) operations were employed to help the model accurately localize and detect targets of various sizes, thereby improving detection accuracy and robustness. Experimental results on a substation equipment infrared image dataset containing three types of devices demonstrate that the improved Faster R-CNN achieved AP values of 92.83%, 94.26%, 96.85%, and 95.97% for current transformers, voltage transformers, insulator strings, and arresters, respectively, with a detection speed of 22 fps. The model can effectively handle challenges such as complex backgrounds, occlusions, and multiscale targets. In particular, noise and brightness tests showed that the improved Faster R-CNN exhibited strong robustness and stability.