Selection of Outline Descriptors Based on LightGBM with Application to Infrared Image Target Recognition

Algorithm 1

Feature selection using LightGBM.

	Input: dataset D, feature set ;
Output: optimal feature subset ;
Step 1: LightGBM is used to calculate the importance of sample feature elements , respectively.
Step 2: the feature elements are arranged in descending order according to the results obtained in the first step.
Step 3: LightGBM is used to evaluate and calculate the accuracy of the current feature subset F.
Step 4: the sorted feature subset is selected in step 2 by deleting the feature element with the least importance in F and calculating the recognition accuracy .
Step 5: if , then go back to step 3 for execution; otherwise, recycle the feature elements deleted in the previous step and then go back to step 3 for execution.
Step 6: when all the feature elements have been traversed, the optimal feature subset is output.