Fine Metal Mask (FMM) is the core technology of the evaporation process (EV) in manufacturing AMOLED panels [1]. The FMM made by the welding process. As the AMOLED's resolution becomes higher, the FMM becomes thinner, and the inspection system for securing the welding reliability becomes an issue in the FMM manufacturing process. Currently, inspection is conducted manually on random samples thus it is impossible to inspect all welds. In this study, the red‐dot labeled image training model was developed to automatically classify and determine weld defects, thereby enhancing the detection consistency of welds. This paper applies weld‐defect‐qualification‐function based on deep learning for FMM mass production process. The ‘Red‐dot labeled image training method' draws a red point in the center of the weld and separates the red‐channels. The separated channels are used as labeled images for learning. The problem is that if the model learns the welds directly, a detection error that catches the wrong area occurs, and the welding area is not correctly cropped. Additionally, the image processing and OpenCV's contour are errors occur by the image's contrast or facility light condition. On the other hand, the weld area can be correctly cropped by the red‐dot training model, even only with 1 pixel. The effect of this technology is to solve the problem of defect leakage and deviation of workers resulting from the existing manual sampling inspection, and to make a consistent qualification and automatic full inspection based on the deep learning model. Through this paper, the problem of full‐automation‐inspection, which was not possible in the past, is solved and the foundation of the inspection technology applicable to mass production is established.