Acta Photonica Sinica, Volume. 50, Issue 11, 1128001(2021)
Remote Sensing Images Target Detection Based on Adjustable Parameter and Receptive field
Figures & Tables(13)
Fig. 6. Visual detection results of ASDN512.For each category，ASDN512 is shown on the left and ASDN32 on the right
Table 1. Number of instances of each category in DIOR
View tableView in ArticleTable 1. Number of instances of each category in DIOR
Category Number Category Number Airplane 10 104 Ground Track Field 3 038 Airport 1 327 Harbor 5 509 Baseball Field 5 817 Overpass 3 114 Basketball Court 3 225 Ship 62 400 Bridge 3 967 Stadium 1 268 Chimney 1 681 Storage Tank 26 414 Dam 1 049 Tennis Court 12 266 Expressway Service Area 2 165 Train Station 1 011 Expressway Toll Station 1 298 Vehicle 40 370 Golf Field 1 086 Wind Mill 5 363
Table 2. Confusion matrix
View tableView in ArticleTable 2. Confusion matrix
Actual Predict Positive Negative Positive True Positive（TP） False Negative（FN） Negative False Positive（FP） True Negative（TN）
Table 3. Comparison of the accuracy and receptive field of the combination of different dilation ratios
View tableView in ArticleTable 3. Comparison of the accuracy and receptive field of the combination of different dilation ratios
Dilation ratios （1，2，3） （1，2，5） （1，3，5） （3，5，7） mAP/% 67.5 67.8 68.1 67.4 Receptive field 188 220 236 332
Table 4. Object classes in the DIOR dataset
View tableView in ArticleTable 4. Object classes in the DIOR dataset
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 Airplane Airport Baseball field Basketball court Bridge Chimney Dam Expressway
service area
Expressway
toll station
Golf field C11 C12 C13 C14 C15 C16 C17 C18 C19 C20 Ground track field Harbor Overpass Ship Stadium Storage tank Tennis court Train station Vehicle Wind mill
Table 5. Comparison of detection accuracy of different algorithms on DIOR
View tableView in ArticleTable 5. Comparison of detection accuracy of different algorithms on DIOR
Model *RICNN［ 21 ］FaterRCNN［ 11 ］*Fater⁃RCNN+FPN［ 22 ］CBD⁃E［ 23 ］ASDN512 （Ours） *YOLO⁃V3［ 7 ］*SSD［ 8 ］*CornerNet
［
24 ］*Retinanet
［
25 ］ASDN32 （Ours） Backbone VGG16 VGG16 ResNet⁃
101
ResNet⁃
101
VGG16 Darknet⁃
53
VGG16 Hourglass⁃104 ResNet‐
101
VGG16 C1 39.1 61.5 54 54.2 65.9 72.2 59.5 58.8 60.2 63.9 C2 61 72.5 74.5 77 73.5 29.2 72.7 84.2 72 73.8 C3 60.1 68.3 63.3 71.5 73.7 74 72.4 72 70.6 71.8 C4 66.3 81.9 80.7 87.1 81.7 78.6 75.7 80.8 80.5 81 C5 25.3 45.1 44.8 44.6 48 31.2 29.7 46.4 43.6 46.3 C6 63.3 73.1 72.5 75.4 73.5 69.7 65.8 75.3 72.3 73.4 C7 41.1 57.6 60 63.5 56.6 26.9 56.6 64.3 61.4 56.3 C8 51.7 70.8 75.6 76.2 75.1 48.6 63.5 81.6 72.1 73.4 C9 36.6 63.9 62.3 65.3 66 54.4 53.1 76.3 66.7 66.2 C10 55.9 74.6 76 79.3 75.9 31.1 65.3 79.5 72 74.7 C11 58.9 79.8 76.8 79.5 77.2 61.1 68.6 79.5 73.4 75.2 C12 43.5 43.5 46.4 47.5 51.7 44.9 49.4 26.1 45.3 51.1 C13 39 59 57.2 59.3 59 49.7 48.1 60.6 56.9 58.4 C14 9.1 47.6 71.8 69.1 76.7 87.4 59.2 37.6 71.7 76.2 C15 61.1 67.9 68.3 69.7 68.2 70.6 61 70.7 70.4 67.4 C16 19.1 42.4 53.8 64.3 63.8 68.7 46.6 45.2 62 60.2 C17 63.5 81.1 81.1 84.5 81.7 87.3 76.3 84 80.9 81.4 C18 46.1 53.5 59.5 59.4 61 29.4 55.1 57.1 57 58.7 C19 11.4 34 43.1 44.7 45.8 48.3 27.4 43 47.2 45.8 C20 31.5 83.3 81.2 83.1 86.3 78.7 65.7 75.9 84.5 83.1 mAP/% 44.2 63.1 65.1 67.8 68.1 57.1 58.6 64.9 66.1 66.9
Table 6. Ablation experiments
View tableView in ArticleTable 6. Ablation experiments
Model K DP RFA CNA AFC mAP% FPS Parameters Faster RCNN × × × × × 59.5 19.6 136.85M FRCNN32 × × × √ × 58.8 34.4 40.76M FRCNN512 × × × √ × 59.4 19.2 139.31M FRCNN512C × × × √ √ 59.5 22.2 75.13M FRCNN K √ × × × × 64.9 19.6 136.85M FRCNN K- √ √ × × × 67.5 18.8 136.85M FRCNNK -+ √ √ √ × × 68.1 18.1 136.85M ASDN32 √ √ √ √ √ 66.9 32.3 21.44M ASDN512 √ √ √ √ √ 68.1 20.4 75.13M
Tools
Get Citation
Copy Citation Text
Nan LIU, Zhaoyong MAO, Yichen WANG, Junge SHEN. Remote Sensing Images Target Detection Based on Adjustable Parameter and Receptive field[J]. Acta Photonica Sinica, 2021, 50(11): 1128001
Paper Information
Category: Remote Sensing and Sensors
Received: May. 7, 2021
Accepted: Jul. 6, 2021
Published Online: Dec. 2, 2021
The Author Email: SHEN Junge (shenjunge@nwpu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20