Mosaic ceramic art pattern with noble, elegant features, it is a unique form of art creation in ancient Greece and the ancient Rome period has been loved by artists and created a lot of classic large-scale exterior mosaic ceramic art works. Small size square mosaic ceramic as the basic raw material for the creation of large exterior mosaic art, it directly affects the quality of the work created by the artist, so these ceramic mosaic ceramic materials need to undergo rigorous inspection to meet the needs of the artist’s high-quality art creation. However, small size multi-color square mosaic ceramics are different from ordinary large target ceramics, they have the characteristics of small size and easy reflection, currently mainly using manual inspection, the existing automatic inspection methods have the problem of low efficiency and accuracy, cannot meet the needs of artists for the quantity and quality of mosaic ceramics. To solve these problems, this paper proposes a new convolutional network-based fast nondestructive testing method for detecting square mosaic tiles. The detection method is based on the convolutional neural network YOLOv5s model, and by introducing the AF-FPN module and the data enhancement module, the method further improves the recognition performance of the model relative to the original YOLOv5s model and achieves the fast detection of surface defects on square mosaic ceramics. The experimental results show that the detection method for small size multicolor square mosaic ceramic tile surface minor defects detection rate of up to 94 % or more, a single square mosaic ceramic detection time of 0.41 s. The method takes into account the detection accuracy and speed, can be fast and accurate screening of high-quality, defect-free small size multicolor square mosaic ceramic, to meet the artist’s requirements for high-quality mosaic ceramic raw materials Quality and quantity requirements, to ensure the quality of the creation of mosaic art patterns, to better show the charm of the mosaic art patterns role. At the same time, the method can not only be applied to the detection of mosaic ceramics, the method can also be applied to have a similar small volume, easy to reflect the characteristics of small target object defect detection.