Automatic Detection of Cabbage Pest Attacks Based on Leaf Images with Machine Learning Approach

Abstract

Early detection of cabbage pest attacks is essential for reducing crop losses and improving agricultural productivity. This paper proposes an automatic cabbage pest detection system based on leaf images by integrating Gray-Level Co-occurrence Matrix (GLCM) texture feature extraction with a Decision Tree Boosting classification approach. The proposed method consists of image acquisition, preprocessing, GLCM feature extraction, and machine learning-based classification to identify pest damage severity. The extracted texture features include contrast, dissimilarity, correlation, homogeneity, angular second moment (ASM), energy, and entropy, which characterize the texture changes caused by pest infestations. Experimental results demonstrate that the proposed approach effectively distinguishes healthy and pest-attacked cabbage leaves and classifies damage into five severity levels: Normal, Low, Moderate, Severe, and Very Severe. The analysis further shows that contrast and entropy increase with increasing damage severity, whereas homogeneity and energy decrease, indicating that GLCM features provide discriminative texture information for pest identification. The Decision Tree Boosting classifier successfully utilizes these features to produce consistent and objective classifications, reducing the reliance on manual visual inspection. Therefore, the proposed framework provides a practical and computationally efficient solution for automatic cabbage pest detection and damage assessment. The developed system has the potential to support precision agriculture by enabling timely pest management decisions, minimizing crop losses, and improving cabbage production. Future research should evaluate the proposed approach using larger field datasets and investigate the integration of advanced deep learning models and multispectral imaging to further improve detection accuracy and robustness.