Enhancing Power Line Infrared Image Segmentation with Pyramid Pooling

This paper addresses a significant challenge in segmenting infrared images of power transmission line equipment, where traditional methods often struggle with environmental interferences and the equipment's complex structures. To overcome these limitations, we introduce the Pyramid Pooling Module (PPM) to enhance segmentation performance. The PPM combines features at multiple scales, improving the model's adaptability to targets of different sizes. This approach not only refines how features are represented but also strengthens the model's ability to distinguish subtle details within complex backgrounds. Experimental results demonstrate that integrating the PPM into the network architecture significantly outperforms other decoder configurations in segmenting infrared images of power transmission lines. Implementing PSPHead achieves the best overall performance, preserving image detail integrity while boosting overall recognition accuracy.