Segmentation Method for Whole Vehicle Wood Detection Based on Improved YOLACT Instance Segmentation Model

In order to overcome the problems of slow detection speed, low accuracy, dense wood stacks and easily obscured overlooked, a segmentation method based on YOLACT_WOOD is proposed.YOLACT algorithm proposed explore feasibility single-stage instance model for fast accurate whole-truck wood. this study, original YOLACT model, firstly, ResNeXt network embedded with CBAM attention mechanism module used as backbone imporve feature extraction capability model; secondly, image input size increased improve ability medium small diameter class wood; then CIoU bounding box regression loss function accuracy regression; finally, DIoU combined Fast-NMS boundary screening problem false missed detections. evaluated using five evaluation metrics: mAP, FPS, IoUmask , true rate, parametric size, mask map fitted counted OpenCV library. The experimental results show that xmlns:xlink="http://www.w3.org/1999/xlink">mAP study improved by 5.6% compared network, xmlns:xlink="http://www.w3.org/1999/xlink">IoUmask 2.6%, xmlns:xlink="http://www.w3.org/1999/xlink">FPS 14.7 frames/sec speed Mask R-CNN rate logs in test set reaches 96.61%, 0.23%, number not significantly improved. This result shows only ensures but also improves solves omission, has strong robustness generalisation ability.