Ensemble-Based Deep Learning for Automated Insect Pests Detection in Black Pepper: A Transfer Learning Approach

Insect pests pose a significant threat to black pepper cultivation, affecting both crop yield and quality. Rapid and accurate identification of these pests is crucial for implementing timely control measures and minimizing economic losses. This study proposes an advanced computer vision approach to automate the detection and classification of pest insects in black pepper crops, addressing the limitations of traditional manual inspection methods. We present a novel ensemble-based deep learning model that leverages transfer learning techniques to achieve high-accuracy pest classification. Our approach combines the strengths of three pretrained convolutional neural networks (CNNs) - VGG16, VGG19, and ResNet50 - integrated through a voting classifier ensemble. This parallel pipeline architecture allows robust feature extraction and improved generalization between diverse pest species. To train and evaluate our model, we utilized the PIBPv1.0 dataset, a comprehensive collection of 2,378 images spanning five distinct classes of black pepper pest insects. This data set was curated from various publicly available sources to ensure diversity and applicability in the real world. The experimental results demonstrate the efficacy of our ensemble approach, achieving an overall classification accuracy of $\mathbf{9 8 3 2 \%}$. This performance highlights the potential of our system to serve as a valuable tool for black pepper farmers and agricultural professionals, enabling rapid identification of insect pests and informing targeted pest management strategies. Our research contributes to the growing field of precision agriculture, offering a scalable and efficient solution to the persistent challenge of pest insect detection in black pepper cultivation.