Knowledge distillation-based high-precision lightweight bridge crack identification method

• Main Authors: FAN QIAN, XIA ZHANGHUA, TAN YUNHUA, ZHU SANFAN
• Format: Patent
• Language: Chinese; English
• Published: 03.11.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention discloses a high-precision lightweight bridge crack identification method based on knowledge distillation. The method mainly comprises the following steps: (1) preparing a data set; and constructing a self-established data set T with a label and a data set XU without a label. And (2) training a teacher network model. And dividing the self-built data set T in the step 1 into a training set, a verification set and a test set in a ratio of 6: 2: 2 to obtain a teacher model with the highest accuracy. And (3) training a student network model. A novel BCSLD (semi-supervised label knowledge distillation) scheme is adopted, a teacher model is used for guiding a student model to learn, and a high-precision lightweight recognition model is constructed. And (4) deploying the model. And deploying the high-precision lightweight recognition model obtained by training to embedded equipment, connecting a camera and arranging the camera on an unmanned aerial vehicle, so that the task of real-time intelligent det