Image classification method and device for continuous test scene and computer equipment

• Main Authors: ZHU CHENG, ZHANG WEIMING, YANG HAO, ZHU XIANQIANG, JIANG JINSHEN, ZHOU YUN
• Format: Patent
• Language: Chinese; English
• Published: 26.03.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to a continuous test scene-oriented image classification method and device and computer equipment. The method comprises the following steps: acquiring a pre-trained image classification model; and inputting a to-be-classified image into the image classification model, and introducing matrix noise into a batch normalization layer in the image classification model to obtain an image classification result. In order to alleviate a possible overfitting problem of an image classification model in a continuous test scene, the invention explores a relationship between disturbance noise and image classification model generalization in a continuous loop test scene, and introduces a portable and effective parameter-free method. Matrix-level disturbance noise is added into batch normalized statistical parameters, so that an image classification task that a source pre-training model adapts to a target domain under domain transfer can be better completed. 本申请涉及一种面向连续测试场景的图像分类方法、装置和计算机设备。所述方法包括：获取预先训练好