基于改进YoloX的输电通道工程车辆检测识别
2022,30(9):67-73
摘要：针对输电通道下环境复杂，各类工程车辆频繁损坏输电线路中所需解决的对工程车辆的检测识别问题，在单阶段目标检测算法YoloX的基础上，对YoloX算法中的损失函数进行修改，平衡正负样本和难易样本，在网络中添加CBAM注意力机制，将内部通道信息和位置信息结合，提高特征的提取能力，并通过修改强特征提取部分Neck中的CspLayer结构，在保证检测速度的前提下，提高模型的检测性能。通过筛选亮度低的图片，引入改进的MSR算法对图片进行亮度提升，优化数据集。实验结果表明，提出的算法提高了检测的准确率，与传统的YoloX算法相比，mAP提高了4.64%，识别效果明显提升，证明了新算法的有效性。
关键词：目标检测;工程车辆;YoloX;注意力机制;MSR
Detection and recognition of transmission channel engineering vehicles based on improved YoloX
Abstract：In view of the complex environment under the transmission channel and the frequent damage of various engineering vehicles to the transmission line, the problem of detection and identification of engineering vehicles needs to be solved. Based on the single-stage target detection algorithm YoloX, the loss function in YoloX algorithm is modified to balance the positive and negative samples and difficult samples, add CBAM attention mechanism in the network, combine the internal channel information and location information, improve the feature extraction ability, and modify the CspLayer structure in the strong feature extraction part Neck, on the premise of ensuring the detection speed, Improve the detection performance of the model. By screening the pictures with low brightness, the improved MSR algorithm is introduced to improve the brightness of the pictures and optimize the data set. Experimental results show that the proposed algorithm improves the detection accuracy. Compared with the traditional YoloX algorithm, the mAP is improved by 4.64%, and the recognition effect is significantly improved, which proves the effectiveness of the new algorithm.
Key words：target detection; engineering vehicles; YoloX; attention mechanism; MSR
收稿日期：2022-03-09
基金项目：国家自然科学基金青年基金资助项目(61903183)
