• Laser & Optoelectronics Progress
  • Vol. 58, Issue 20, 2028002 (2021)
Zeming Li1, Liang Cheng2、3、4、5, Daming Zhu1、*, Zhaojin Yan2、3, Chen Ji2、3, Zhixin Duan2、3, Min Jing2, Ning Li2, Shengkun Dongye1, Yanruo Song1, and Jiahui Liu6
Author Affiliations
  • 1Faculty of Land and Resources Engineering, Kunming University of Science and Technology, Kunming, Yunnan 650093, China
  • 2School of Geography and Ocean Science, Nanjing University, Nanjing, Jiangsu 210023, China
  • 3Collaborative Innovation Center of South China Sea Studies, Nanjing, Jiangsu 210023, China
  • 4Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing, Jiangsu 210023, China
  • 5Jiangsu Center for Collaborative Innovation in Novel Software Technology and Industrialization, Nanjing, Jiangsu 210023, China
  • 6School of Geography and Ecotourism, Southwest Forestry University, Kunming, Yunnan 650051, China
  • show less
    DOI: 10.3788/LOP202158.2028002 Cite this Article Set citation alerts
    Zeming Li, Liang Cheng, Daming Zhu, Zhaojin Yan, Chen Ji, Zhixin Duan, Min Jing, Ning Li, Shengkun Dongye, Yanruo Song, Jiahui Liu. Deep Learning and Spatial Analysis Based Port Detection[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2028002 Copy Citation Text show less
    Structural diagram of YOLO v3 network
    Fig. 1. Structural diagram of YOLO v3 network
    Wharf marking at different datasets. (a) DIOR; (b) TGRS-HRRSD; (c)(d) level 19 Google remote sensing image
    Fig. 2. Wharf marking at different datasets. (a) DIOR; (b) TGRS-HRRSD; (c)(d) level 19 Google remote sensing image
    PR curve of wharf recognition
    Fig. 3. PR curve of wharf recognition
    Recognition results of wharves. (a) Original image; (b) local image; (c) recognition of wharf with single ship docked; (d) recognition of wharf with many ships docked; (e) wharf recognition in complex scene; (f) wharf recognition when prescene of flares on sea surface; (g) recognition of jetty wharf; (h) recognition of along-shore wharf; (i)(j) typical misrecognitions
    Fig. 4. Recognition results of wharves. (a) Original image; (b) local image; (c) recognition of wharf with single ship docked; (d) recognition of wharf with many ships docked; (e) wharf recognition in complex scene; (f) wharf recognition when prescene of flares on sea surface; (g) recognition of jetty wharf; (h) recognition of along-shore wharf; (i)(j) typical misrecognitions
    (a) Port hotspots and (b)--(e) aggregated polygons
    Fig. 5. (a) Port hotspots and (b)--(e) aggregated polygons
    Scale of feature mapOriginal anchor boxesNew anchor boxes
    52×5210×13; 16×30; 33×2331×97; 31×30; 47×156
    26×2630×61; 62×45; 59×11957×49; 85×93; 130×43
    13×13116×90; 156×198; 373×326164×162; 293×300; 511×511
    Table 1. Comparsion of anchor boxes before and after adjustment
    ModelATPAFPAFNprF1
    Original YOLO v31258756358.96%18.17%27.78%
    Proposed algorithm36413732472.65%52.91%61.23%
    Table 2. Comparison of performances between two models in experimental area
    Threshold /mNumber of hotspotsNumber of portsProportion of ports in hotspotsNumber of aggregated polygons
    5002241174878.00%438
    8001932146875.98%442
    10001757135377.01%465
    12001617126177.98%450
    15001481115778.12%435
    Table 3. Hotspot analysis results under different thresholds
    Zeming Li, Liang Cheng, Daming Zhu, Zhaojin Yan, Chen Ji, Zhixin Duan, Min Jing, Ning Li, Shengkun Dongye, Yanruo Song, Jiahui Liu. Deep Learning and Spatial Analysis Based Port Detection[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2028002
    Download Citation