Lightweight DeepLabv3+ Segmentation Method for Substation Equipment Oil Leakage Image

To efficiently and accurately detect and identify oil leakage in the substations oil-filled equipments, this paper proposes a lightweight segmentation model based on the improved DeepLabv3+ model. By using the lightweight segmentation model based on the improved DeepLabv3+ model. By using the lightweight MobilenetV2 network as the backbone feature extraction network, the model reduces the number of network parameters and improves the semantic segmentation efficiency. To improve the segmentation accuracy of small leaked oil region and the edges, a feature pyramid structure is introduced to enhance the feature extraction ability. In addition, a dataset of oil leakage images from substation oil-filled equipments is constructed for the oil leakage image segmentation, and experimental tests show that the proposed algorithm significantly improves segmentation performance. Compared with the DeepLabv3+ algorithm-based oil leakage detection model, the average intersection-over-union (IoU) is improved by 1.28%, and the number of model parameters decreased to 15.5% of the original. It meets the practical requirements of edge deployment model, and has high practicality.