一种改进DeepLabV3+的岩屑图像语义分割算法

An improved cuttings image semantic segmentation algorithm based on DeepLabV3+

岩屑图像的准确分割与识别是地质勘探人员研究地层特征和地质建模的重要基础和保障。针对DeepLabV3+在岩屑图像语义分割时存在的推理速度慢、分割效果不理想的问题,提出了一种改进的Deep⁃LabV3+岩屑图像语义分割模型。首先,利用轻量级MobileNetV3提取岩屑图像语义特征,以减少参数量和计算量;其次,采用结合注意力机制的空洞空间卷积池化金字塔模块(atrous spatial pyramid pooling,ASPP)以更好地拟合岩屑图像特征;然后,借鉴特征金字塔思想将编码过程提取的不同尺度特征融合至解码过程,从而获取更精准的岩屑图像语义分割结果;最后,在自制的岩屑图像数据集上和现有的7种方法进行对比实验,实验结果表明,改进的DeepLabV3+模型的类别平均像素准确率为0.78,平均交并比为0.68,相比于其他的网络模型具有更好的分割表现。

Accurate segmentation and recognition of cuttings image is an important basis and guarantee for geological explor⁃ers to study stratigraphic characteristics and geological modeling.An improved DeepLabV3+cuttings image semantic segmentation model is proposed to solve the problems of slow reasoning speed and unsatisfactory segmentation effect in DeepLabV3+cuttings im⁃age semantic segmentation model.Firstly,lightweight MobileNetV3 is used to extract the semantic features of cuttings images to re⁃duce the number of parameters and computational cost.Secondly,Atrous spatial pyramid pooling(ASPP)module combined with attention mechanism is used to better fit the cuttings image features.Then,the different scale features extracted from the encoding process are fused to the decoding process by using the feature pyramid idea,so as to obtain more accurate semantic segmentation re⁃sults of cuttings images.Finally,compared with the existing 7 methods on the self⁃made cuttings image dataset,the experimental re⁃sults show that the improved DeepLabV3+model has a category average pixel accuracy of 78%and an average intersection ratio of 68%.Compared with other network models,it has better segmentation performance.