Abstract: Detection of cracking sound in duck eggshells is an important aspect of the egg processing industry. In this paper, we propose a method for detecting cracking sounds in duck eggs by utilizing multi-feature extraction and optimizing the light gradient boosting machine (LightGBM) classification model using improved grey wolf optimization (IGWO). The proposed method aims to address the issues of low accuracy and slow detection caused by inconsistent thickness of duck eggshells. It involves using principal component analysis (PCA) to select time-frequency domain features from the crack signal, extracting eight feature vectors, optimizing the parameters of the LightGBM classification model through IGWO algorithm, and evaluating its classification performance. Experimental results demonstrate that compared to using all features as input, employing selected features can improve detection speed by two times on a collected dataset of cracked duck eggs. Moreover, with IGWO-LightGBM, the recognition accuracy for duck eggs reaches 96.64%, which is 15.02 percentage points higher than that achieved by support vector machine (SVM). This method significantly enhances both recognition accuracy and detection speed, making it suitable for industrial assembly lines.