• Acta Optica Sinica
  • Vol. 39, Issue 6, 0615001 (2019)
Jiangrong Xie1、2、3, Fanming Li1、3、*, Hong Wei1, Bing Li1, and Baotai Shao1、2、3
Author Affiliations
  • 1 Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China
  • 2 University of Chinese Academy of Sciences, Beijing 100049, China
  • 3 Key Laboratory of Infrared System Detection and Imaging Technology, Chinese Academy of Sciences, Shanghai 200083, China
  • show less
    DOI: 10.3788/AOS201939.0615001 Cite this Article Set citation alerts
    Jiangrong Xie, Fanming Li, Hong Wei, Bing Li, Baotai Shao. Enhancement of Single Shot Multibox Detector for Aerial Infrared Target Detection[J]. Acta Optica Sinica, 2019, 39(6): 0615001 Copy Citation Text show less
    Structure of SSD network
    Fig. 1. Structure of SSD network
    Schematics of multiple feature fusion methods. (a) Pooling; (b) transposed deconvolution; (c) bi-direction fusion
    Fig. 2. Schematics of multiple feature fusion methods. (a) Pooling; (b) transposed deconvolution; (c) bi-direction fusion
    Diagram of semantic segmentation branch
    Fig. 3. Diagram of semantic segmentation branch
    Comparison of detection results of small targets obtained by original SSD and improved SSD. (a) Original SSD; (b) improved model
    Fig. 4. Comparison of detection results of small targets obtained by original SSD and improved SSD. (a) Original SSD; (b) improved model
    Detection results of infrared aerial targets
    Fig. 5. Detection results of infrared aerial targets
    Comparison of recall-precision curve of infrared aerial targets
    Fig. 6. Comparison of recall-precision curve of infrared aerial targets
    Convolutional layerConvolutional receptive field /(pixel×pixel)Output scale of feature layer /(pixel×pixel)Default boxes ratioMapping region scale /(pixel×pixel)
    conv4_392×9238×380.1030×30
    conv7276×27619×190.2060×60
    conv8_2340×34010×100.37111×111
    conv9_2468×4685×50.54162×162
    conv10_2724×7243×30.71213×213
    conv11_2980×9801×10.88264×264
    Table 1. Convolution receptive field and mapping image region of default boxes of SSD_300×300
    MethodmAPDetection result
    Aero planeBirdBoatBottleCarDogSheepPerson
    SSD_300×3000.5370.6010.5250.4260.3740.7200.5560.5380.563
    Proposed method0.6080.6850.5700.5340.5030.7480.5970.6520.591
    Table 2. Small object detection results of VOC2007 dataset
    MethodmAPDetection result
    Fighter_JHelicopterFighter_SAirlinerBird
    SSD_300×3000.6180.6590.6150.2660.8190.732
    YOLOv3-3200.6410.7300.5480.3140.8670.747
    Proposed method0.7050.7840.6360.4850.8220.796
    Table 3. Aerial target detection results of infrared dataset
    MethodNumber of boxesTotal boxesFPS
    75×7538×3819×1910×105×53×31×1
    Original SSD0466644873225.2
    Modified SSD4444444302609.4
    Table 4. Number of predictive boxes for each classi?ed network and speed comparison
    Jiangrong Xie, Fanming Li, Hong Wei, Bing Li, Baotai Shao. Enhancement of Single Shot Multibox Detector for Aerial Infrared Target Detection[J]. Acta Optica Sinica, 2019, 39(6): 0615001
    Download Citation