【摘要】 YOLOV4 Tiny目标检测算法是通过卷积神经网络提取特征，进行预测类别和边界框坐标的经典深度学习算法，是YOLOV4目标检测算法的简化版，没有使用Mish激活函数来提取特征，而只使用特征金字塔来增强特征层，因此不需要进行下采样。存在的不足是检测精度比较低。文章针对YOLOV4 Tiny算法存在的不足进行了改进，将低层特征层与高层特征层进行特征融合，然后分别进行三次空洞卷积，在扩大感受野的同时也能捕获多尺度上下文信息，并将结果进行堆叠，取代原网络中的FPN特征金字塔。实验结果表明，改进后的YOLOV4 Tiny算法比原算法精度更高，满足实时要求，具有一定程度的鲁棒性。更多还原

【Abstract】 YOLOV4 Tiny target detection algorithm is a classical deep learning algorithm that extracts features through reel neural networks, predicts categories and bounding box coordinates, is a simplified version of YOLOV4 target detection algorithm, does not use Mish activation function in feature extraction, and uses only one feature pyramid in the feature strengthening layer, and no further sampling. The shortcoming is that the detection accuracy is relatively low. In this paper,the shortcomings of the YOLOV4 Tiny algorithm are improved, the low-level feature layer and the high-level feature layer are feature fusion, and then three empty converges are carried out, which can capture multi-scale context information while expanding the feeling field, and stack the results to replace the FPN feature pyramid in the original network. The experimental results show that the accuracy of the improved YOLOV4 Tiny algorithm is much higher than that of the original algorithm,which meets the requirements of real-time performance and has some robustness.更多还原