An Ore Image Segmentation Method Based on Attention-Unet-GAN

In the milling process, ore particle size is an important indicator that affects the efficiency and stability of the mill, accurate and real-time detection of ore particle size is significant for the optimal control of the mill. In the task of ore particle size detection using machine vision methods, ore image segmentation is the key step. However, ore image segmentation is difficult due to the irregular shape of the ore and the extreme similarity of the texture of the ore region and the conveyor belt region in the ore image. Existing CNN-based methods suffer from the difficulty of obtaining global information as well as low segmentation accuracy. This paper proposes a novel ore segmentation method based on Attention-Unet-GAN using Generative Adversarial Training Methods and semantic segmentation network combined with the attention mechanism. The experimental results show that the evaluation metrics, such as segmentation accuracy, IoU and F1-score, are higher than traditional convolutional neural networks, which proves the effectiveness of this method.