| 摘要: |
| 采用加强特征提取网络为MobileNetV2的融合多特征金字塔场景解析网络（PSPnet）来实现复杂场景下的图像语义分割.相对于深度残差网络ResNet50和MobileNetV1，引入了线性瓶颈结构和反向残差结构，利用金字塔池化模块（PPM）来处理不同层级的图像特征信息，并将其进行特征拼接，有效避免了不同分割尺寸下，子区域之间关键特征信息的缺失.在此基础上，引入注意力机制模块，结合通道注意力机制（CAM）和空间注意力机制（SAM），进一步提高分割精度.实验结果表明：该方法可以提高图像识别的准确率，并节省训练时间. |
| 关键词: 语义分割 金字塔池化模块(PPM) 注意力机制 特征融合 |
| DOI：10.3969/J.ISSN.1000-5137.2023.02.003 |
| 分类号:TP391.4 |
| 基金项目: |
|
| Multi-objective semantic segmentation based on PSPnet with attention mechanisms |
|
ZHANG Shuai, YANG Chunxia
|
|
College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
|
| Abstract: |
| A pyramid scene parsing network (PSPNet) with MobileNetV2 as enhanced feature extraction network was adopted to achieve semantic segmentation of images in complex scenes. Compared with the deep residual network ResNet50 and MobileNetV1, linear bottlenecks and inverted residuals were introduced and the pyramid pooling module (PPM) was used to process the image feature information of different layers and to feature stitching, which could avoid missing key feature information between sub-regions under different segmentation sizes effectively. On this basis, the attention mechanism module was introduced to further improve the segmentation accuracy by combining the channel attention mechanism(CAM) with the spatial attention mechanism(SAM). The experimental results verified that the method could achieve the expected goals, improve the accuracy of image recognition and reduce the training time. |
| Key words: semantic segmentation pyramid pooling module (PPM) attention mechanisms feature integration |
