Neural network model test sufficiency evaluation method based on depth operator

The invention discloses a neural network model test sufficiency evaluation method based on a depth operator, and the method comprises the steps: constructing a test data set, and collecting test output according to the operation result of a to-be-tested model; traversing all combinations of the other two labels different from the current real label, and calculating projection points of the original output to the currently selected boundary triangle plane area; secondly, dividing a boundary triangle generated by the selected label group according to a preset recursion round number; judging which areas are covered according to intersection information of projection points in all the current boundary triangular areas and the divided areas; and finally, summarizing all correct label classes and all boundary triangles, and performing normalization calculation after summarizing test sufficiency indexes of the current to-be-evaluated depth operator data set about the to-be-tested neural network model. According to t