Early Detection of Crop Pest Infestation Using Hybrid HOG and VGG-16 Feature Fusion

Abstract

Pest infestation poses a significant threat to agricultural productivity, often going undetected in early stages due to subtle symptoms on crop leaves. Timely and accurate detection is crucial for minimizing crop loss. This study introduces a feature fusion-based deep learning model for automatic pest infestation detection using leaf images. The model combines handcrafted Histogram of Oriented Gradients (HOG) features with deep features extracted from VGG-16 network, capturing both fine-grained textures and high-level semantics. By combining fine-grained texture descriptors with high-level visual representations, the model achieves robust classification of pest-infested leaf images. The dataset consists of RGB images of infected crop leaves, collected under consistent environmental conditions. The proposed model achieves a classification accuracy of 95%, outperforming traditional single-feature and single-model approaches. This image-based, dual-feature strategy offers a scalable and effective solution for early-stage pest detection in precision agriculture.