• Laser & Optoelectronics Progress
  • Vol. 58, Issue 16, 1615007 (2021)
Yuhuan Li1、*, Jie Wang1, Li Lu1, and Ying Nie2
Author Affiliations
  • 1Air Defense and Missile Academy, Air Force Engineering University, Xi'an, Shaanxi 710051, China
  • 2Unit 93861 of Sanyuan, Shaanxi, Xianyang, Shaanxi 713800, China
  • show less
    DOI: 10.3788/LOP202158.1615007 Cite this Article Set citation alerts
    Yuhuan Li, Jie Wang, Li Lu, Ying Nie. Lightweight Real-Time Target Detection Model for Remote Sensing Images[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1615007 Copy Citation Text show less
    Tiny YOLO-V3 network structure
    Fig. 1. Tiny YOLO-V3 network structure
    Improved backbone network structure
    Fig. 2. Improved backbone network structure
    Structure of detection head in Tiny YOLO-V3
    Fig. 3. Structure of detection head in Tiny YOLO-V3
    Improved detection head structure
    Fig. 4. Improved detection head structure
    Partial sample images in dataset. (a) Aircraft; (b) oiltank; (c) flyover; (d) playground
    Fig. 5. Partial sample images in dataset. (a) Aircraft; (b) oiltank; (c) flyover; (d) playground
    Experimental process
    Fig. 6. Experimental process
    Change curve of learning rate in training process
    Fig. 7. Change curve of learning rate in training process
    Change curve of loss value during training
    Fig. 8. Change curve of loss value during training
    Partial test results of different models. (a) Tiny YOLO - V3; (b) test results of improved model; (c) dataset labels
    Fig. 9. Partial test results of different models. (a) Tiny YOLO - V3; (b) test results of improved model; (c) dataset labels
    DatasetAircraftOiltankFlyoverPlayground
    Training set390136141130
    Test set56293519
    Table 1. Number of partitions in dataset
    FacilitieVersion
    Operating system of trainingWindows10.0
    Processor of trainingIntel(R)Core(TM) i7-9700 CPU 3.00 GHz
    Operating system of testLinux raspberrypi 4.19.118-v7l+
    Processor of testCortex-A72(ARM v8)
    PyThon3.6.5
    TensorFlow1.14.0
    Table 2. Configuration of experimental environment
    ModelAPmAP
    AircraftOiltankFlyoverPlayground
    Tiny YOLO-V30.6080.3970.0070.0230.253
    Improved model0.6350.4310.0120.0220.275
    Table 3. mAP values of Tiny YOLO-V3 and improved model
    ModelTime complexitySpace complexityFPS
    Tiny YOLO-V32767598080591762521.83
    Improved model423320000190237083.01
    Table 4. Complexity and frame rate of different models
    Yuhuan Li, Jie Wang, Li Lu, Ying Nie. Lightweight Real-Time Target Detection Model for Remote Sensing Images[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1615007
    Download Citation