复杂大交通场景弱小目标检测技术

针对现有基于大数据和深度学习的目标检测框架对于高分辨率复杂大场景中低分辨率小目标识别效果较差、多目标检测的精度和实时性难以平衡的问题，改进了基于深度学习的目标检测框架 SSD（single shot multibox detector），提出一种改进的多目标检测框架DRZ-SSD，将其专用于复杂大交通场景多目标检测。检测以从粗到细的策略进行，分别训练一个低分辨率粗略检测器和一个高分辨率精细检测器，对高分辨率图像进行下采样获得低分辨率版本，设计了一种基于增强学习的动态区域放大网络框架（DRZN），动态放大低分辨率弱小目标区域至高分辨率再使用精细检测器进行检测识别，剩余图像区域使用粗略检测器进行检测，对弱小目标的检测与识别精度以及运算效率的提高效果明显；采用模糊阈值法调整自适应阈值策略在避免适应数据集的同时提高模型的决策能力，显著降低了检测漏警率和虚警率。实验表明，改进后的DRZ-SSD在应对弱小目标、多目标、杂乱背景、遮挡等检测难度较大的情况时，均能获得较好的效果。通过在指定数据集上的测试，相比于其他基于深度学习的目标检测框架，各类目标识别的平均准确率提高了4%~15%，平均准确率均值提高了约9%~16%，多目标检测率提高了13%~34%，检测识别速率达到38 fps，实现了算法精度与运行速率的平衡。

Detection of dim and small targets in complex large traffic scenes

Aiming at the problems that the existing target detection framework based on big data and depth learning has poor recognition effect on low-resolution small targets in high-resolution complex large-field scenes, and the accuracy and real-time performance of multi-target detection are difficult to balance, this paper improved the SSD based on depth learning, and proposed an improved multi-target detection framework DRZ(dynamic region zoom-in) -SSD, which was dedicated to multi-target detection in complex large traffic scenes. It carried out the detection in a coarse-to-fine strategy. It trained a low-resolution coarse detector and a high-resolution fine detector respectively, downsampled the high-resolution image to obtain a low-resolution version, and designed a dynamic region zoom-in network based on enhanced learning. It dynamically enlarged the low-resolution small target region to a high-resolution and used the fine detector to carry out detection and identification, and detected the remaining image region by using the coarse detector, so that the detection and identification accuracy of the small target and the operation efficiency were obviously improved. It adopted fuzzy threshold method to adjust the adaptive threshold strategy to avoid adapting to the data set and improved the decision-making ability of the model and significantly reduced the detection missed alarm rate and false alarm rate. Experiments show that the improved DRZ-SSD can achieve good results when dealing with weak targets, multi-targets, cluttered background, occlusion and other difficult detection situations. Through testing on the specified data set, compared with other target detection frameworks based on deep learning, the average accuracy rate of various types of target recognition has increased by 4%~15%, the average accuracy rate has increased by 9%~16%, the multi-target detection rate has increased by 13%~34%, and the detection and recognition rate has reached 38 fps, realizing the balance between the accuracy of the algorithm and the running rate.