Tobacco leaf image fine-grained classification method based on soft threshold weighting and context clustering graph convolution

• Main Authors: CHEN XIANYONG, WANG YANGEN, LIAN CHANGWEI, GONG TAO, CHEN QUAN, CHEN FEI
• Format: Patent
• Language: Chinese; English
• Published: 01.03.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to a tobacco leaf image fine-grained classification method based on soft threshold weighting and context clustering graph convolution. The method comprises the following steps: S1, extracting a multi-layer token feature graph from a tobacco leaf image by using a Swindow-Transfomer network added with an FPN (Fabry-Perot Network); s2, performing soft threshold weighting on the token feature map by using a K-means clustering algorithm and a Gaussian function; s3, superimposing attention distribution of multiple layers of token feature maps, and dividing the last layer of token feature map into a main token feature map and a secondary token feature map by using learnable parameters; s4, aggregating contextual information of the main token feature graph, fusing the contextual information into the original feature graph, regarding the token features as nodes, and establishing a graph according to the similarity of the nodes; and S5, aggregating all information through the graph convolutional n