Expansive soil crack image recognition method based on convolutional neural network

• Main Authors: CONG SHENGYI, ZHANG XIYANG, MAO XIAOGANG, BAE YUN-SANG, CHEN HONGWEI, ZHANG ZHONGYUAN, LING XIANCHANG, LI XINYU, CHENG ZHIHE, TANG LIANG
• Format: Patent
• Language: Chinese; English
• Published: 25.10.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention discloses an expansive soil crack image recognition method based on a convolutional neural network, and the method comprises the following steps: 1, collecting an expansive soil sample, and carrying out the CT scanning of the expansive soil sample, and obtaining a CT image; step 2, processing the CT image obtained in the step 1 to obtain a binarized image; 3, labeling the binarized images, grouping the binarized images, and establishing a sample set 1 and a sample set 2; step 4, establishing a convolutional neural network model and training the convolutional neural network model by using the sample set 1; 5, performing accuracy evaluation on the convolutional neural network model trained in the step 4 by using the sample set 2; if the difference between the two is not more than 2%, the convolutional neural network model is successfully trained, otherwise, a random inactivation layer is added for re-training; and 6, identifying the expansive soil crack picture by using the trained convolutional n