The mainstream detection and classification methods of traditional advanced ceramic parts are pure mechanical size filtering and manual judgment. To solve the problems of high cost， high error rate and high damage rate， a multi-object real-time detection and classification model （MRDC） based on deep learning is proposed. The model is based on YOLOv3， uses SKNet as an attention mechanism for feature reconstruction to improve accuracy， and cooperates with grayscale map fast transformation algorithm and frame skipping detection method to improve detection speed， which can realize real-time defect detection. Image data in multiple batches are collected， and each batch of data contains 1 000 images of multiple ceramic parts. The mean average precision reaches 99.19% when the advanced ceramic parts in actual production are collected and trained. The correct rate of recognition and classification reaches 100% when the production line video of advanced ceramic parts is used for inspection， which can guarantee to detect 450—550 objects per minute. The MRDC model has the advantages of faster recognition， higher recognition accuracy， and less damage to objects， which can greatly save production raw materials and labor costs， reduce scrap output， and protect the environment more.