Progress in remote sensing image recognition 2022/5/5

1. Select some tags to use as training sets

Take the proportion of labels in the current box as the indicator of selection .

2. Select all tags to use as the training set

3. Consider image clustering to select labels

The clustering effect is too bad , Not adopted .

4. Reduce the resolution of the data set to train

Consider whether the resolution is too high , Leading to poor feature extraction .

5. experimental result

6. Use feature extraction to judge label quality

Here are just the categories of an image . Find out the commonalities of all image categories （ feature extraction ）, Then the outlier detection is used to determine whether the original label is labeled correctly .

problem ： In this way, we can only judge what is correct in the original label , It is not possible to judge what is not marked in the background （ The training mainly cuts the original image in the form of sliding window , Therefore, the content in the background will also participate in the training .）