Image classification method and system based on superpixel segmentation

• Main Authors: YIN BENJUN, MAO RUNXIN, WANG YUANFENG, SUN JIALONG
• Format: Patent
• Language: Chinese; English
• Published: 13.06.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention provides an image classification method and system based on superpixel segmentation. The method comprises the following steps: carrying out superpixel segmentation on an image; converting the superpixel segmentation area into word vectors; and inputting the word vectors into the trained Transform neural network model to carry out classification and identification on the image. Compared with the input of a regular patch with a fixed size, the super-pixel serving as visual word is more similar to the understanding of a natural language on human information, the semantic information of the image is more completely reserved, the method can be analogous to the modeling of a Transform on the natural language, the model calculation performance is improved, and the accuracy of an image classification result is higher. 本发明提供了一种基于超像素分割的图像分类方法及系统，所述方法包括：对图像进行超像素分割；将超像素分割区域转化成词向量；将词向量输入训练好的Transformer神经网络模型对图像进行分类识别。本发明的优势在于：相比于规则固定大小patch的输入，本发明所提出的超像素作为visual word更加类似于自然语言对人类信息的理解，同时图像的语义信息保留更加完整，可以类比于Tra