ENVI supervised classification error: resolution is 0.00000 solution

question

Recently, we need to conduct supervised classification of images. After selecting the region of interest, an error occurred when calculating the separation. The separability of all samples was 0, as shown in the figure.

So I searched on Baidu and saw that many people said it was because the ROI samples were too few or too large or too small. So I deleted the samples again and selected as many as possible, and tried to keep the sizes consistent, but this error still occurred.
At this time, I saw this article: [http://zhihu.geoscene.cn/article/2158] , you can also refer to it

Solution

In retrospect, there may be something wrong with my image itself. The value range of each band is different, resulting in "overflow".
So I checked the range of each band and found that the range of band1-band3 is between 0-225, but the value of band4 is all 255, so I only need to remove band4 and keep band1-band3 to classify .

Next, I will talk about how to eliminate a certain band from the image.

How to delete a certain band from an image using ENVI

First download envi app store (note: only supports ENVI 5.3 and higher, and needs to be connected to the Internet)
Download address: https://envi.geoscene.cn/appstore/
After downloading, decompress the compressed package and get this Two files

. Copy the first file to the extensions under the envi installation path. In my case, it is D:\ENVI53\ENVI53\extensions. After

opening envi, the app store will appear in the Extensions in the toolbox.

Double-click to open the app store and search. "Split multi-band images into multiple single-band files", click Install. After successful installation, you need to restart ENVI. After restarting, the tool Split to Multiple Single-Band Files will appear in the extension. Double-click and select the files that need to be split. file, click OK, set the output format and path to split the image into bands.