基于PReNet和YOLOv4融合的雨天交通目标检测网络

为提高恶劣雨天交通环境下车辆目标检测精度,提出一种基于PReNet和YOLOv4融合的深度学习网络DTOD-PReYOLOv4,融合了改进的图像复原子网D-PReNet和改进的目标检测子网TOD-YOLOv4;将多尺度膨胀卷积融合模块和添加了挤压激励块的注意机制残差模块引入PReNet,获得的D-PReNet可更有效提取雨纹特征; 使用轻量化的CSPDarknet26代替YOLOv4骨干模块CSPDarknet53,为YOLOv4的颈部路径聚合网络模块添加复合残差块,同时采用k-means++算法取代原始网络聚类算法,获得的TOD-YOLOv4可在改善交通小目标检测精度的同时进一步提高检测效率; 基于构建的雨天交通场景车辆目标数据集VOD-RTE对DTOD-PReYOLOv4进行了验证。研究结果表明：与当前YOLO系列主流网络相比,提出的DTOD-PReYOLOv4对原浅层ResBlock_body1叠加残差块,可以更好地提取分辨率较小的特征; 对原深层ResBlock_body3、ResBlock_body4和ResBlock_body5进行裁剪,获得ResBlock_body3×2、ResBlock_body4×2和ResBlock_body5×2,可以有效降低卷积层冗余,提高内存利用率; 为PANet中的Concat+Conv×5添加跳跃连接形成CRB模块,可以有效缓解网络层数加深引起的小目标检测效果退化问题; 采用k-means++算法,在多尺度检测过程中为较大的特征图分配更加适合的较小先验框,为较小的特征图分配更加适合的较大先验框,进一步提高了目标检测的精度; 与MYOLOv4相比,精确率和召回率的调和平均值、平均精度、检测速度分别提升了5.02%、6.70%、15.63帧·s^-1,与TOD-YOLOv4相比,分别提升了3.51%、4.31%、2.17帧·s^-1,与YOLOv3相比,分别提升了46.07%、48.05%、18.97帧·s^-1,与YOLOv4相比,分别提升了31.06%、29.74%、16.26帧·s^-1。     

Technique based on image pyramid and bayes rule for noise reduction in unsupervised change detection

In this paper, a technique based on image pyramid and Bayes rule for reducing noise effects in unsupervised change detection is proposed. By using Gaussian pyramid to process two multitemporal images respectively, two image pyramids are constructed. The difference pyramid images are obtained by point-by-point subtraction between the same level images of the two image pyramids. By resizing all difference pyramid images to the size of the original multitemporal image and then making product operator among them, a map being similar to the difference image is obtained. The difference image is generated by point-by-point subtraction between the two multitemporal images directly. At last, the Bayes rule is used to distinguish the changed pixels. Both synthetic and real data sets are used to evaluate the performance of the proposed technique. Experimental results show that the map from the proposed technique is more robust to noise than the difference image.