AN EFFICIENT YOLOV9 MODEL FOR CAR PARTS DETECTION AND SEGMENTATION

Assessing car damages from an accident is a crucial process in the car insurance industry. At present, this task requires a manual inspection of each component. However, it is anticipated that smart devices will perform these evaluations more efficiently in the future. To address this, our research introduces an efficient YOLOv9 model featuring two novel modules. The RepNCSPELAN4 module splits input from initial convolutional layer into two paths and is processed through RepNCSP and convolutional layers before merging. This dual-path strategy enhances gradient flow and feature reuse, improving learning efficiency. The SPPELAN module utilizes the convolutional layer for channel adjustment and is followed by spatial pooling operations to capture multi-scale contextual information. This concatenates outputs and consolidates features through another convolutional layer, optimizing detailed feature extraction across spatial hierarchies. Experimental validation on the car parts dataset demonstrates superior performance, with our model achieving higher mAP (67.7%) on object detection and mAP (65.8%) on segmentation detection.