基于Faster-RCNN与自注意力机制的矿山图像异常检测算法

Mining Image Anomaly Detection Algorithm Based on Faster-RCNN and Self-attention Mechanism

矿山异常行为检测是一项重要工作,有助于提高采场安全生产监管效率。提出了一种基于自注意力机制的Faster-RCNN算法,用于矿山图像异常行为检测。该算法通过自注意力机制对特征图进行加权,有效提取关键特征,并减少冗余信息干扰。首先,从图像中提取RoI区域,并通过自注意力机制对RoI区域内的特征图进行加权,使得关键特征得到更准确的提取。同时,对RoI区域内的特征图进行细粒度融合,以捕捉更多的图像细节信息。最后,使用分类器对每个RoI区域进行分类,以检测图像中的异常行为。在自建数据集上对算法进行了训练与评估,试验结果表明:所提算法在矿工图像异常行为检测工作上表现出更高的准确性和鲁棒性,与传统的Faster-RCNN算法相比,该算法检测精度提高了4.8%。此外,该算法对于光照和角度等变化具有更好的鲁棒性,可以有效应对实际场景中的复杂环境。

Mine abnormal behavior detection is an important task,which can help for improving the supervision efficiency for mine safety production.A Faster-RCNN algorithm based on self-attention mechanism is proposed to detect abnormal behavior in mine images.The algorithm can extract key features effectively and reduce the interference of redundant information by weighting feature maps with self-attention mechanism.Firstly,the RoI region is extracted from the image,and the feature map in the RoI region is weighted by the self-attention mechanism,so that the key features can be extracted more accurately.At the same time,the feature maps in the RoI region are fused to capture more image details.Finally,a classifier is used to classify each RoI region to detect abnormal behavior in the image.The algorithm is trained and evaluated on the self-built data set,and the test results show that the proposed algorithm has higher accuracy and robustness in the abnormal behavior detection task of miners,and the detection accuracy of the proposed algorithm is improved by 4.8%compared with the traditional Faster-RCNN algorithm.In addition,the algorithm has better robustness to changes in illumination and angle,and can effectively deal with complex environments in real scenes.