Figure 5 – Administrator interface for managing user privileges 5.1.3 Creation of dataset and generation of classification model The application provides an efficient and intuitive workflow for the entire process, from the creation of training images to the generation of discriminant models and their application to unknown images. The detailed procedure is outlined below. (1) Preparation of Training Dataset The initial stage of model building requires the creation of a training dataset. As illustrated in the Figure 6(a), the user uploads a .dat file of hyperspectral images from a PC to the application. Following this, the user is required to label the image. The application permits the user to establish up to three distinct labels; for instance, "Component A," "Component B," and "Other" are anticipated to be classified. The user is then able to select any area in the image and assign a corresponding label using a mouse drag operation, as illustrated in the Figure 6(b). This operation is intuitive and greatly improves the work efficiency of labeling compared to the conventional method. The labelled images are stored as a training data set and used in the subsequent model generation process. (2) Discriminant Model Generation Upon completion of the step, the user is then able to initiate the building of a discriminant model by clicking the "Generate Model" button, as illustrated in the Figure 6(c). During this process, the status on the application changes to "Generating" and "Generated" is displayed upon completion of the model construction. The model can be generated again using the same procedure to create a discrimination model that reflects the latest information, should new images be added to the data set or existing labels be modified. (3) Application of the decision model to unknown images, the discriminant model that has been generated can also be used to classify unknown hyperspectral images. The user uploads the image to be judged to the application and applies the discriminant model to it.
RkJQdWJsaXNoZXIy MTM0Mzk2