基于RetinaNet的煤矿井下输送带异物检测技术

Conveyor Belt Foreign Object Detection Technology Based on RetinaNet in Underground Coal Mine

受煤矿生产环境与开采条件等因素影响,煤炭中常混入煤矸石、铁器等异物,从而在井下煤炭运输过程中对输送带造成损害,导致经济损失和安全隐患。使用深度学习方法对输送带异物进行检测。在煤矿井下光照和粉尘影响下,运用HSV空间改进融合Retinex算法对图像进行增强,以RetinaNet_Res101为基础网络,用八度卷积代替网络中的部分传统卷积层,在特征提取阶段降低特征图中低频分量冗余特征,提高细节特征提取效果,减少空间冗余,达到提升精度的同时节约计算资源提高运算速度的效果。实验结果表明,八度卷积优化RetinaNet模型在测试集平均精度为94.1%,比原始RetinaNet模型提高3.9%,同时检测速度提高了26.3%。

Affected by factors such as the production environment and mining conditions of coal mine,coal gangue, iron and other foreign objects are often mixed into the coal, which will cause damage to the belt during the underground coal transportation process, resulting in economic losses and safety hazards. The deep learning method was used to detect the foreign objects of conveyor belt. Under the influence of light and dust in underground coal mine, the image was enhanced by using the HSV space improvement and fusion Retinex algorithm. Taking the RetinaNet_Res101 as basic network, the octave convolution was used to replace some traditional convolution layers in the network. In the feature extraction stage, reduced the redundant features of low-frequency components in the feature map,improved the effect of detail feature extraction, reduced spatial redundancy, and achieved the effect of improving accuracy while saving computing resources and improving computing speed. The experimental results show that the average accuracy of the octave convolution optimized RetinaNet model in the test set is 94.1%, which is 3.9% higher than the original RetinaNet model, and the detection speed is increased by 26.3%.