In this study, we propose the deep learning based platform for defect classification based on machine vision and perform the empirical analysis with various image data supplied by the mid or small-sized companies in South Korea to evaluate the performance of the proposed approach. Through this platform, we aim to perform full inspection of products with reasonable cost investment. Specifically, this study uses deep learning based algorithm to perform foreign material inspection, processing shape inspection, and processing omission inspection based on product image data of three manufacturing companies. A classification model and anomaly detection model were used, and empirical analysis was performed through actual product image data. All defective products were judged as defective, and a small number of good products were judged as false defects, with an accuracy of 97.9%. Through the experiments, we believe that a machine vision-based defect classification model we developed would be a highly cost-effective way to identify defects through images. Through this manufacturing data-based platform for the same industry, it is expected that even small and medium-sized enterprises with low manufacturing data-based technology can introduce and develop defect classification technology at a low cost.