基于注意力机制的道路环境语义分割算法

Semantic Segmentation Algorithm of Road Environment Based on Attention Mechanism

针对复杂场景道路图像分割中,由于目标形状不规则、光照变化以及物体遮挡等因素,而导致的分割结果出现分割精度低、小目标分割错误等问题,提出了一种新的语义分割算法GH-deeplabV3+。将DeeplabV3+网络和高分辨率网络相结合,并在骨干特征提取网络中插入注意力机制模块。高分辨率网络模块保持了图像的高分辨特征信息,注意力机制模块增强了关键目标特征信息的权重。在优化超参数的研究中,使用APReLU激活函数和AdaBelief优化器来优化算法,降低网络损失。在Cityscapes数据集上进行了验证,实验结果表明,GH-deeplabV3+算法提高了图片的分割精度,分割性能优于其它分割算法。

In the road image segmentation of complex scenes,owing to factors such as irregular target shape,illumination changes and object occlusion,the segmentation results have low segmentation accuracy and small target segmentation errors.A new semantic segmentation algorithm CH-deeplabV3+is proposed,we combine the DeeplabV3+and the high-resolution network,and insert the attention mechanism module into the backbone feature extraction network.The high-resolution network module maintains the high-resolution feature information of the image,and the attention mechanism module enhances the weight of the key target feature information.In the study of optimizing hyperparameters,the APReLU activation function and AdaBelief optimizer are used to optimize the algorithm and reduce network loss.It has been verified on the Cityscapes dataset.The experimental results show that the GH-deeplabV3+algorithm improves the segmentation accuracy of the image,and the segmentation performance is better than that of other segmentation algorithms.