Improving Semantic Segmentation of Switch Connector Images in DeepLabV3+

The projected area of the switch connector can reflect the product quality of the component. In order to accurately determine the quality of the switch connector, it is necessary to calculate the projected area of the product. This article proposes a switch connector image segmentation model based on an improved DeepLabV3+ network to achieve precise segmentation of switch connectors in images. This network uses the MobileNetV2 network pre trained using the PASCAL VOC public dataset as the backbone network for feature extraction of images, and then uses this network to connect with the improved DeepLabV3+ network as the final segmentation network. The improved DeepLabV3+ network uses the sum of Dice Loss and cross entropy loss function on the basic DeepLabV3+ network to replace the original method that only uses one cross entropy loss function, so as to alleviate the adverse effects caused by the imbalance between foreground and background in the samples; At the same time, ECA module is also added as an attention mechanism module to enhance the feature representation of the model, thereby improving the segmentation performance of the model. The experiment shows that the segmentation results of this model reach 95.17% in average intersection to union ratio (MIOU) which is 1.30 percentage points higher than the original DeepLabV3+ model, and the frame rate is 12.04% higher. At the same time, the network model in this paper also reduces the parameter quantity and computational cost by 89.38% and 83.58% compared to the original model, respectively. The method proposed in this article can effectively perform product quality inspection on industrial parts with multiple sizes, types, complex shapes and characteristics in complex industrial scenarios, achieving automated and intelligent quality inspection for factory production, while improving quality inspection efficiency and reducing labor costs.