A Lightweight Bilateral Semantic Segmentation Network for Robots

Agricultural robots have experienced rapid development in the past few years, and real-time autonomous moving is considered to be their basic capability. In order to improve the robot’s autonomous transportation performance, a lightweight bilateral semantic segmentation network for agricultural robots is proposed. Firstly, the improved bilateral semantic segmentation network is proposed to realize the segmentation with high precision and efficiency through the aggregation layer. Secondly, a PID adaptive control algorithm is used to control the steering angle of the agricultural robot. Finally, in order to verify the effectiveness of the improved semantic segmentation network, MIoU, the average intersection ratio frequently utilized in semantic segmentation, is used to evaluate the segmentation accuracy. Moreover, the proposed network is used to compare with Fast-SCNN, DeeplabV3+, BiSeNetV1, and BiSeNetV2 algorithms using the cityscape dataset. Experimental results show that the proposed network achieves more accurate segmentation than the above methods and is successfully applied to the agricultural robot in an outdoor environment.