Training method for multi-output land cover classification model, classification method, and device

A training method for multi-output land cover classification model, comprising:

• obtaining a training data;
• inputting the training data into an initial model based on deep belief nets for training to obtain a multi-output land cover classification model, wherein the initial model comprises N level outputs, and the N level outputs comprise an output set at the last network layer of the initial model and (N−1) level output set at any (N−1) network layers from a first network layer to a penultimate network layer of the initial model;
• determining a total loss according to losses of the N level outputs;
• performing a backpropagation based on the total loss to adjust a parameter of the initial model;
• wherein the step of obtaining the training data comprises:
• obtaining an image set for training;
• performing a feature extraction on the image set to obtain the training data, wherein the extracted features comprise at least one of a band spectrum, a vegetation index, a principal component band, a multi-scale filtering feature, a multi-scale texture feature, and a terrain feature;
• wherein N=2, and the initial model comprises a first level output and a second level output;
• the step of determining the total loss according to the losses of the N level outputs comprises:
• obtaining a first loss based on the first level output and a corresponding label, and obtaining a second loss based on the second level output and a corresponding label;
• and
• calculating the total loss based on the first loss, the second loss, and corresponding weight coefficients of the first loss and the second loss.