改进的Faster R-CNN在车辆识别中的应用

Application of improved Faster R-CNN in vehicle recognition

随着人工智能技术的不断发展,利用深度学习进行车辆识别已经成为智能交通领域的热点。以更快速区域卷积神经网络(Faster R-CNN)模型为基础,利用BIT-Vehicle数据集定义车辆视觉任务,利用改进的困难样本算法加强对图像中远小目标车辆的特征提取,并改进NMS算法的置信度函数,动态调整置信度区间,对实际场景中的车辆图像进行测试。该方法可输入多尺度图像、能自主提取车辆特征,提高模型对小目标的判别能力,减少漏检情况,提升检测精度,加速模型收敛,鲁棒性较强。

With the continuous development of artificial intelligence technology,the application of deep learning for vehicle recognition has become a hotspot in the field of intelligent transportation. On the basis of the faster regions with convolutional neural network(Faster R-CNN)model,the BIT-Vehicle dataset is adopted to define vehicles visual task,the improved hard example algorithm is utilized to enhance the feature extraction of the far small target vehicles in the image,the confidence function of the NMS algorithm is improved,the confidence interval is adjusted dynamically,and the vehicle image in the actual scene is tested. The method can input multi-scale images and automatically extract vehicle features,improve the ability of the model to discriminate small targets, reduce the missed detection, improve the detection accuracy, and accelerate the convergence of the model. The method also has strong robustness.