Lightweight RGB-T transmission line detection method and equipment based on knowledge distillation and comparative learning guidance

Through the method based on knowledge distillation and comparative learning guidance, the problems of high calculation cost and low recognition accuracy in existing RGB-T transmission line detection are solved, and efficient and accurate transmission line detection is achieved, reducing the number of parameters and enhancing the model robustness and generalization ability.

CN118941814APending Publication Date: 2024-11-12ENJOYOR COMPANY LIMITED
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410949600.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-12

Smart Images

  • Figure CN118941814A_ABST
    Figure CN118941814A_ABST
Patent Text Reader

Abstract

The invention discloses a lightweight RGB-T transmission line detection method and device based on knowledge distillation and comparative learning guidance, and belongs to the technical field of transmission line detection of deep learning. The method comprises the following steps: constructing and training a double-flow teacher convolutional neural network to obtain a weight of the double-flow teacher convolutional neural network; constructing a uniflow student convolutional neural network; constructing a training framework based on knowledge distillation and comparative learning guidance, and training a single-flow student convolutional neural network based on the training framework and the weight of the double-flow teacher convolutional neural network; based on the trained double-flow teacher convolutional neural network and single-flow student convolutional neural network, obtaining a knowledge distillation and comparative learning guided lightweight RGB-T transmission line detection network model; and obtaining a to-be-detected transmission line image, and performing prediction through the knowledge distillation and comparative learning guided lightweight RGB-T transmission line detection network model to obtain a saliency prediction image.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Cited By

  • Road segmentation method based on multi-modal feature adaptive progressive fusion

    CN119399478A

  • Automatic driving simulation scene dynamic iteration method and system for large model closed-loop distillation

    CN122509041A