• Laser & Optoelectronics Progress
  • Vol. 56, Issue 6, 061007 (2019)
Bo Xie*, Bin Zhu, Hongwei Zhang, Qi Ma, and Yang Zhang
Author Affiliations
  • State Key Laboratory of Pulsed Power Laser Technology, National University of Defense Technology, Hefei, Anhui 230037, China
  • show less
    DOI: 10.3788/LOP56.061007 Cite this Article Set citation alerts
    Bo Xie, Bin Zhu, Hongwei Zhang, Qi Ma, Yang Zhang. Gradient Clustering Algorithm Based on Deep Learning Aerial Image Detection[J]. Laser & Optoelectronics Progress, 2019, 56(6): 061007 Copy Citation Text show less
    Network structure of SSD object detection algorithm
    Fig. 1. Network structure of SSD object detection algorithm
    Structural diagram of algorithm
    Fig. 2. Structural diagram of algorithm
    Schematic of map division in STING algorithm. (a) Rendering of grid raster on map; (b) rendering of clustering on map
    Fig. 3. Schematic of map division in STING algorithm. (a) Rendering of grid raster on map; (b) rendering of clustering on map
    Test results
    Fig. 4. Test results
    Relationship between frame rate and image size of SSD based on sliding window method
    Fig. 5. Relationship between frame rate and image size of SSD based on sliding window method
    Relationship between running time and clustering data in gradient clustering algorithm
    Fig. 6. Relationship between running time and clustering data in gradient clustering algorithm
    Frame rate versus image size and data efficiency in gradient clustering algorithm
    Fig. 7. Frame rate versus image size and data efficiency in gradient clustering algorithm
    Frame rate versus image pixels and dispersion in SSD algorithm based on gradient clustering
    Fig. 8. Frame rate versus image pixels and dispersion in SSD algorithm based on gradient clustering
    Recall rates for different object categories
    Fig. 9. Recall rates for different object categories
    Relationship between object recall rate and normalized dispersion
    Fig. 10. Relationship between object recall rate and normalized dispersion
    AlgorithmKernel sizeSigmaGrid sizeDensity
    Gaussian filter7×71.5--
    STING--300600
    Table 1. Related parameter settings in proposed algorithm
    MethodV^FPS /s-1mAP /%AP /%
    PlaneShipStoragetankGroundtrack fieldHarborBridge
    Faster R-CNN0.1951.3192.5548.8368.010.559.9138.10
    YOLO0.5536.5963.6175.5680.3500.050
    SW-SSD0.3448.8082.4684.3876.650.538.5410.29
    GC-SSD0.6546.9670.1479.3174.720.536.9920.08
    Table 2. Detection results based on DOTA dataset
    Bo Xie, Bin Zhu, Hongwei Zhang, Qi Ma, Yang Zhang. Gradient Clustering Algorithm Based on Deep Learning Aerial Image Detection[J]. Laser & Optoelectronics Progress, 2019, 56(6): 061007
    Download Citation