基于轻量化网络和注意力机制的智能车快速目标识别方法

为提高智能车在真实环境中的实时检测能力，改善复杂环境下检测效果不佳的问题，本文提出一种基于轻量化网络和注意力机制的智能车快速目标识别方法。首先，为了减少网络计算参数和提升目标识别算法的推理速度，提出利用GhostNet加速YOLOv4的特征提取；其次，为了提高复杂场景下对道路目标的识别精度，在GhostNet和特征金字塔部分添加结合软阈值化改进的注意力模块；最后，为了验证本文提出方法的有效性，选取Pascal VOC、KITTI公开数据集和自制城市道路数据集进行实验对比。与其他目标检测算法在精度和速度上进行比较，结果证明，本文方法在平均检测精度提升1.7%的情况下，模型参数量降低到原来的18.7%，检测速度提升了 66%，检测速度和精度均优于其他算法，可满足智能车的实时感知需求。

Intelligent Vehicle Target Fast Recognition Based on Lightweight Network and Attention Mechanism

To improve the real-time detection ability of intelligent vehicle in real environment and improve the detection effect in complex environment, this paper proposes a target fast recognition method of intelligent vehicle based on lightweight network and attention mechanism. The GhostNet is proposed to accelerate the feature extraction of YOLOv4, which also helps to reduce the network calculation parameters and improve the reasoning speed of the target recognition algorithm. Then, an improved attention module combined with soft thresholding is added to GhostNet and feature pyramid to improve the recognition accuracy of road targets in complex scenarios. The Pascal VOC, KITTI public data sets and some assumed urban road datasets are selected for experimental comparison to verify the effectiveness of the proposed method. Compared with other target detection algorithms in terms of accuracy and speed, the average detection accuracy of this method is increased by 1.7%, the model parameters are reduced to 18.7% of the original, and the detection speed is increased by 66% . The detection speed and accuracy from the proposed method are higher than the traditions algorithms, which can meet the real-time perception needs of intelligent vehicles.