基于改进YOLO v3的机械装置目标检测算法

Object Detection Algorithm of Mechanical Equipment Based on Improved YOLO v3

基于计算机视觉的机械装置检测技术虽然已有研究,但大多检测效率低。为此提出基于改进的YOLO v3算法的机械装置识别方法,对YOLO v3算法进行剪枝优化研究,即通过对图像数据标注,构建机械装置红外图像数据集;运用L1正则化将神经网络稀疏化进而剔除冗余通道,以达到网络精简。实验结果表明,进行剪枝后的目标检测算法综合性能有较大提升,其模型体积缩小82.14%,运行速度加快74.38%,准确率为92.47%,检测速率为27.9 fps.通过性能对比分析,证实此方法在满足检测准确率条件下显著提高了检测速率,降低检测所使用设备的硬件要求及功耗,易于满足实际中对机械装置快速识别检测的要求。

Although the mechanical device detection technology based on computer vision has been studied,most of these methods have the problem of low detection efficiency.An identification method of mechanical device based on the improved YOLO v3 algorithm is proposed to study pruning optimization of YOLO v3 algorithm.The infrared image data set of mechanical device is constructed by labeling the image data.L1 regularization is used to sparse the neural network,and then redundant channels are eliminated to achieve network simplification.The experimental results show that the comprehensive performance of the object detection algorithm after pruning has been greatly improved.The model volume is reduced by 82.14%,the operating speed is accelerated by 74.38%,the accuracy rate is 92.47%,and the detection rate is 27.9 fps.By comparing and analyzing the performance of the model,it is proved that the method proposed in this paper not only satisfies the detection accuracy rate,but also significantly improves detection speed,and reduces the hardware requirements and power consumption of the equipment used in the detection,and it is easy to meet the requirements of rapid detection of mechanical devices in practice.