Spacecraft Recovery & Remote Sensing, Volume. 45, Issue 1, 123(2024)
Extraction of Bare Rock Information in Rocky Desertification Area Based on Improved DeepLabV3+
Figures & Tables(11)
![[in Chinese]](/Images/icon/loading.gif){kind=icon}
Table 1. MobileNetV3 network structure
View tableView in ArticleTable 1. MobileNetV3 network structure
特征层 输入特征 操作 SE 激活函数 步距 1 3×5122 Conv2d — HS 2 2 16×2562 Bneck, 3×3 — RE 1 16×2562 Bneck, 3×3 — RE 2 3 24×1282 Bneck, 3×3 — RE 1 24×1282 Bneck, 5×5 YES RE 2 4 40×642 Bneck, 5×5 YES RE 1 40×642 Bneck, 5×5 YES RE 1 40×642 Bneck, 3×3 — HS 2 5 80×322 Bneck, 3×3 — HS 1 80×322 Bneck, 3×3 — HS 1 80×322 Bneck, 3×3 — HS 1 80×322 Bneck, 3×3 YES HS 1 6 112×322 Bneck, 3×3 YES HS 1 112×322 Bneck, 5×5 YES HS 2 7 160×162 Bneck, 5×5 YES HS 1 160×162 Bneck, 5×5 YES HS 1 160×162 Con2d, 1×1 — HS 1 8 960×162 pool, 7×7 — — 1 960×12 Con2d, 1×1, NBN — HS 1 9 1 280×12 Con2d, 1×1, NBN — — 1
Table 2. CA-DC-MobileNetV3 network structure
View tableView in ArticleTable 2. CA-DC-MobileNetV3 network structure
特征层 输入 操作 CA 激活函数 步距 空洞率 1 3×5122 Conv2d — HS 2 1 2 16×2562 Bneck, 3×3 — RE 1 1 16×2562 Bneck, 3×3 — RE 2 1 3 24×1282 Bneck, 3×3 — RE 1 1 24×1282 Bneck, 5×5 YES RE 2 1 4 40×642 Bneck, 5×5 YES RE 1 1 40×642 Bneck, 5×5 YES RE 1 1 40×642 Bneck, 3×3 — HS 1 1 5 80×642 Bneck, 3×3 — HS 1 2 80×642 Bneck, 3×3 — HS 1 2 80×642 Bneck, 3×3 — HS 1 2 80×642 Bneck, 3×3 YES HS 1 2 6 112×642 Bneck, 3×3 YES HS 1 2 112×642 Bneck, 5×5 YES HS 1 2 7 160×642 Bneck, 5×5 YES HS 1 2 160×642 Bneck, 5×5 YES HS 1 2 160×642 Con2d, 1×1 — HS 1 1
Table 3. Performance comparison of different improvement strategies
View tableView in ArticleTable 3. Performance comparison of different improvement strategies
实验 特征提取网络 FPN CA 异感受野特征融合 IOU/% F1/% 参数量 MobileNetV3 CA-DC-MobileNetV3 1 √ × × × × 69.07 81.71 5.63×106 2 × √ × × × 69.99 82.35 4.41×106 3 × √ √ × × 71.49 83.38 4.44×106 4 × √ √ √ × 71.85 83.62 4.44×106 5 × √ √ × √ 72.02 83.73 4.50×106 6 × √ √ √ √ 72.46 84.03 4.51×106
Table 4. Performance comparison of different algorithms
View tableView in ArticleTable 4. Performance comparison of different algorithms
提取方法 IOU/% F1/% 参数量 面向对象 33.55 50.25 — PSPNet 52.54 68.88 4.67×107 SegNet 67.23 80.40 2.94×107 DeepLabV3+ 67.84 80.84 5.69×107 SegFormer MiT-b0 68.45 81.27 0.37×107 本文提取方法 72.46 84.03 0.45×107
Table 5. Comparison of experimental results
View tableView in ArticleTable 5. Comparison of experimental results
模型 场景1 场景2 场景3 场景4 原始影像 标签 面向对象 PSPNet SegNet DeepLabV3+ SegFormer MiT-b0 本文提取方法
Tools
Get Citation
Copy Citation Text
Yongjun WU, Hong WANG, Chen YANG. Extraction of Bare Rock Information in Rocky Desertification Area Based on Improved DeepLabV3+[J]. Spacecraft Recovery & Remote Sensing, 2024, 45(1): 123
Paper Information
Category:
Received: Mar. 20, 2023
Accepted: Jun. 1, 2023
Published Online: Apr. 22, 2024
The Author Email: WANG Hong (7653606@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20