Real-Time Detection Method for Cracked Eggs Based on Improved YOLOv8 and Dual Verification With Triple Classification Granularity

Addressing the urgent need for real-time detection of cracked eggs on production lines, this study proposes a high-precision real-time detection algorithm based on an improved YOLOv8 framework. This study innovatively introduced a dual-detection method, incorporating a “crack region” classification to assist in distinguishing between intact eggs and cracked eggs. This method identifies cracked eggs by recognizing either local or overall crack characteristics, effectively reducing both false negatives and false positives. At the architectural level, the model innovatively integrates Weighted Bidirectional Feature Pyramid (BiFPN) to optimize multi-scale feature fusion, Deformable Convolutional Network v2 (DCNv2) to enhance perception of irregular geometric deformations in cracks, Weighted Intersection over Union v2 (WIoUv2) loss function to improve detection accuracy for occluded samples, and adds a small object detection layer to strengthen recognition of minute defects. Experiments demonstrate that the improved model achieves a mean average precision (mAP) of 92.9% on the test set, representing a 5.2% improvement over the original model and outperforming current mainstream detection models. In real-world production deployment testing, the system achieved a real-time detection speed of 45 FPS, with a false negative rate of only 0.46% and a false positive rate of 2.18% for intact eggs, while keeping the false positive rate for broken eggs below 2%. This performance far surpasses that of the original model and fully meets the requirements for industrial-grade real-time online quality inspection. This study provides an efficient solution for egg quality inspection with significant engineering application value.