SMR-RS: An Improved Mask R-CNN Specialized for Rolled Rice Stubble Row Segmentation

As a highly productive rice, ratoon rice is widely planted worldwide, but the rolling of stubble in mechanical harvesting severely limits its total yield; based on this, some scholars have proposed rolled righting machines. However, limited by uncertainty field environment, machine’s localization accuracy target needs to be improved. To address this problem, real-time detection rows prerequisite. Therefore, paper introduces deep learning method for first time achieve this. end, we presented novel approach improve model that used simplification Mask R-CNN, which does not require any modules added or replaced original model. Firstly, two branches second stage were deleted, and region proposals output from was directly as mask generation region, segmentation performance substantially improved after simple optimization proposals. Further, contribution feature map counted, backbone network simplified accordingly. The resulting SMR-RS still able perform instance has better than R-CNN other state-of-the-art models while significantly reducing average image processing hardware consumption.