PAG-Net:Semantic Segmentation Network of 3D Point Clouds in Excavator Operation Scene based on Attention Mechanics and Ghost Convolutional

Accelerating the construction of intelligent mining platforms has become an essential path to achieving high-quality development in the industry. Currently, there are still several challenges in developing mine intelligence, including the limited intelligence of mining excavators, particularly their struggle to recognize different operation scenarios. In recent years, with the rapid development of artificial intelligence and deep learning, more and more semantic segmentation methods have been proposed. However, most of the current semantic segmentation methods have a problem: how to possibly achieve higher accuracy with fewer parameters and shorter time. To solve this problem, this paper proposes a novel semantic segmentation approach of 3D point clouds in excavator operation scenes, which integrates the attention mechanism with the lightweight convolutional module. We proposed a semantic segmentation network of 3D point clouds in excavator operation scenes based on attention mechanics and ghost convolutional, called PAG-Net. It includes an input embedding module, a light attention module, and ghost convolution modules. Among them, the input embedding module can transform point clouds into a high-dimensional feature space, retaining more feature information. The light attention module can better focus on the features of the point clouds while balancing efficiency and accuracy. And the ghost convolution modules can accelerate the speed of convolution without losing too much information. Our experimental results show that, in the semantic segmentation task of excavator operation scenes, our method is faster and uses fewer parameters compared to other methods, while also achieving better accuracy.