<p>Friction stir welding (FSW) leads to formation of steel flashes at the aluminum substrate which highly affect bo5th corrosion and other properties of the welds. This study is based on machine learning based detection and classification of corrosion mechanisms and investigating the effect of steel flashes, and microstructures on corrosion mechanisms of FSWed joints of AA6061-T6 and AISI304 steel. The FSW leads to accumulation of steel fragments (53–210&#xa0;µm) in the AA6061-T6 substrate which leads to inhomogeneity in corrosion mechanisms. The size and distributions of the steel flashes in the weld nugget have been quantified by using image processing approach. Different types of image thresholding methods have been used such as Binary, Binary-inverse, Adaptive and Otsu thresholding methods. Among four thresholding methods, Binary and Otus’s thresholding provided accurate object detection. Steel fragments increased the corrosion susceptibility of the welds and heterogeneity of the welds. Finer grains, steel flashes and precipitates Mg₂Si (β″) and Al₄Mg₈Si₇Cu₂ (Q) in the aluminum substrate led to environmental, pitting and inter-granular modes of corrosion on it whereas steel substrate was unaffected from the corrosive medium. Corrosion mechanisms were predicted and classified with over 90% accuracy using YOLOv7 and YOLOv7X models. The YOLOv7X model outperformed YOLOv7, with its E-ELAN architecture achieving a mean average precision (mAP) of 83%, compared to 74% for YOLOv7’s ELAN architecture.</p>