A Quality Evaluation Method of UAV Inspection Images Based on Semantic Segmentation

The widespread application of unmanned aerial vehicle (UAV) in power line inspection has resulted in the collection of a large volume of images. While many researchers focus on utilizing deep learning techniques to process these images, there has been limited research on image quality assessment. Existing studies on UAV inspection images face several challenges, including overexposure, degrading and deviant shooting angles which not capture critical components effectively. To evaluate the quality of inspection images better, firstly we definite 7 common major parts mainly shoot by UAV in power transmission lines: tower base, tower body, tower head, cross-arm, wire connection plate, insulator and wire clamp. Meanwhile, the DeepLabv3+ model is employed for semantic segmentation and utilize Densenet121 as the backbone network. Transfer learning and fine-tuning are employed to enhance the performance of the backbone network. Based on the segmented results, every segmented part is evaluated and the overall evaluation result is derived through the weighted composite calculation of evaluation result of all segmented parts. With the proposed method, the evaluation result is closer to the actual needs, focusing on the power line targets. Experimental results demonstrate the effectiveness of the proposed method in identifying UAV inspection images and this research contributes to the field of image quality assessment in UAV-based power line inspection.