• Laser & Optoelectronics Progress
  • Vol. 56, Issue 22, 222803 (2019)
Zexing Du*, Jinyong Yin, and Jian Yang
Author Affiliations
  • Computer Division of Jiangsu Automation Research Institution, Lianyungang, Jiangsu 222002, China
  • show less
    DOI: 10.3788/LOP56.222803 Cite this Article Set citation alerts
    Zexing Du, Jinyong Yin, Jian Yang. Remote Sensing Image Detection Based on Dense Connected Networks[J]. Laser & Optoelectronics Progress, 2019, 56(22): 222803 Copy Citation Text show less
    Design of expanding block. (a) Structure of expanding block; (b) receptive field of expanding block
    Fig. 1. Design of expanding block. (a) Structure of expanding block; (b) receptive field of expanding block
    Structure of dense connected network
    Fig. 2. Structure of dense connected network
    Structure of network
    Fig. 3. Structure of network
    Structure of pre-trained network
    Fig. 4. Structure of pre-trained network
    Divide the target into large, medium, and small sizes
    Fig. 5. Divide the target into large, medium, and small sizes
    Partial detection results
    Fig. 6. Partial detection results
    Area(0,322)[322,962](962,∞)
    Classessmallmediumlarge
    Percentage /%283240
    Table 1. Object size division standard
    Num-blockLarge /%Medium /%Small /%mAP /%Time /s
    384.5880.2878.4381.480.014
    483.6779.3475.9080.110.010
    584.2557.4150.6566.250.008
    Table 2. Effect of number of dense blocks on detection results under the same network depth
    Growth-rateDepthParams /106Time /smAP /%
    12400.10.01080.11
    121000.60.02185.99
    24400.60.02085.24
    241002.40.07086.59
    4019022.60.23487.72
    Table 3. Experimental results obtained by changing the number of feature layers and network depth when number of dense block is 4
    AlgorithmXTPXFPXFNPR
    SSD52119435270.900.85
    YOLO v356126943760.930.89
    Ours-4056526393900.940.90
    Ours-10060134092590.960.93
    Table 4. Comparison of detection results of different algorithms
    AlgorithmmAPlarge /%mAPmedium /%mAPsmall /%mAP /%Time /s
    Densenet-4082.3778.1264.9276.120.008
    Ours-4083.6779.3475.9080.110.010
    Table 5. Improvement effect of network
    Zexing Du, Jinyong Yin, Jian Yang. Remote Sensing Image Detection Based on Dense Connected Networks[J]. Laser & Optoelectronics Progress, 2019, 56(22): 222803
    Download Citation