Causal interpretation method and device based on image classification, equipment and storage medium

Abstract

The invention relates to an artificial intelligence technology, and discloses a causal interpretation method based on image classification, which comprises the following steps: carrying out initial partitioning on a to-be-analyzed image based on superpixels and a linear iterative clustering algorithm to obtain a plurality of partitioned images; calculating responsibility degrees of the plurality of partition images respectively, and performing secondary partition on the partition images of which the responsibility degrees are greater than a responsibility degree threshold value to obtain a plurality of secondary partition images; the responsibility degrees of the secondary partition images are calculated respectively, and when the number of pixel points of the secondary partition images is smaller than or equal to a pixel point threshold value, or the responsibility degrees of the secondary partition images are equal, the secondary partition images serve as a standard attribution map; and performing causal analysis on the standard attribution graph to obtain a causal analysis result. In addition, the invention also relates to a block chain technology, and the partition image can be stored in a node of a block chain. The invention further provides a causal interpretation device based on image classification, electronic equipment and a storage medium. The accuracy of causal interpretation based on image classification can be improved.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, and in particular, to a causal interpretation method, device, device and storage medium based on image classification. Background technique [0002] Causal interpretation is a direction that has attracted much attention in the field of neural network interpretability methods in recent years. For image classification tasks, the reason is the pixel, and the effect is the classification result. The causal interpretation method measures the relationship between each pixel and the classification result. The strength of causality to achieve attribution for image classification. [0003] The existing causal interpretation methods for image classification usually quantify the causal relationship through the degree of responsibility of the pixel points, use image partitions to replace a single pixel point, and determine the causal interpretation of image classification according to...

AI Technical Summary

Problems solved by technology

However, the disadvantage of this method is that the image partition adopts a simple rectangular partition. Since the main body of the image is often not in a regular shape, it takes a long time to refine and iterate starting from the rectangular partition. At the same time, a large number of samples are required to satisfy the randomness of the partition. Leads to low accuracy in causal interpretation of image classification

Images