Fig 1.
Network structure of SSD model.
Fig 2.
Shufflenet_V2 network structure.
Fig 3.
Shufflenet_V2 network main block (a) DownSampling block (b) Normal block.
Table 1.
The architecture of improved Shufflenet_v2 network.
Fig 4.
Schematic diagram of the position of real Box B and prediction box BGT.
Fig 5.
Improved SSD_Shufflenet_V2 network structure.
Fig 6.
Loss calculation.
Fig 7.
Belt conveyor experimental platform.
Table 2.
The training HyperParameters adopted.
Fig 8.
LablImg software annotates images.
Fig 9.
The confusion matrices: (a) the models of enhanced Shufflenet_v2; (b) SqueezeNet model; (c)Densenet model; (d)MobileNet_V2 model.
Table 3.
Comparison and verification of various lightweight convolutional neural networks.
Fig 10.
The object detection diagram: (a) the models of this paper; (b) without the CIoU model; (c) SSD model.
Fig 11.
Accuracy rate curve of the method proposed in this paper.
Table 4.
Performance comparison between other methods and our model.
Fig 12.
Video target recognition verification of this model.