Data cleansing methods, data cleansing programs, and data cleansing equipment

摘要

PROBLEM TO BE SOLVED: To easily acquire high-quality learning data necessary for learning AI used for image classification. SOLUTION: A process of dividing an image set and over-learning a CNN (convolutional neural network) using a part of the divided image set, and an image classification result using the over-learned CNN. The process of deleting / reclassifying is repeated while changing the composition of the learning image. Then, it is determined whether or not the image set in which the image is deleted / reclassified meets the pass criteria, and the image set is repeatedly subdivided until the pass criteria are met, overfitting the CNN and correcting the image set (image). Delete and reclassify) is repeated. By such processing, data cleansing is advanced while verifying the validity of the image set for learning, and the quality of the image set for learning is automatically improved without human intervention. [Selection diagram] Fig. 6