Development of an AI-based Pest Detection System for Crop Monitoring

Thesis Abstract

The proliferation of pest infestations poses a significant threat to crop yields and food security in modern agriculture, necessitating timely and accurate detection methods to mitigate losses. Traditional pest monitoring techniques rely heavily on manual scouting, which is labor-intensive, time-consuming, and often prone to human error, limiting their effectiveness in large-scale farming contexts. This study aims to develop a robust artificial intelligence (AI)-based pest detection system that leverages deep learning techniques to enhance crop monitoring efficiency. The specific objectives are to design and train convolutional neural networks (CNNs) for pest image classification, evaluate the system's accuracy and real-time detection capabilities, and assess its usability in practical farming environments. Employing a quantitative research design, this study collected a comprehensive dataset comprising 10,000 labeled images of common crop pests, such as aphids, locusts, and caterpillars, captured under various lighting and environmental conditions. The images were sourced from agricultural research centers, local farms, and publicly available repositories. A stratified sampling technique ensured balanced representation of pest categories. The primary data collection instrument was a digital camera integrated with a drone platform, enabling rapid image acquisition across different field zones. The images were subsequently annotated and preprocessed using standard computer vision techniques. The core analytical approach involved training multiple convolutional neural network architectures, including ResNet-50 and InceptionV3, utilizing transfer learning to improve model performance. Model validation was conducted through k-fold cross-validation (k=5), and performance metrics such as accuracy, precision, recall, and F1-score were computed to evaluate classification efficacy. A confusion matrix analysis identified common misclassifications, informing further model refinement. To assess real-time detection capabilities, the system was deployed on edge devices, and latency measurements were obtained. Additionally, usability testing with a sample of 30 farmers and agricultural technicians employed thematic analysis to capture qualitative insights on system adoption and operational challenges. The anticipated findings include the AI system achieving a detection accuracy exceeding 90%, with classification precision and recall rates similarly high across pest categories. The real-time deployment is expected to demonstrate processing latencies below two seconds per image, facilitating prompt decision-making. The usability assessment is projected to reveal favorable perceptions of the system's practicality, alongside areas requiring interface improvements. These results will confirm the potential of AI-driven approaches in revolutionizing pest management practices by enabling early detection and targeted intervention. This research contributes significantly to the body of knowledge by demonstrating the feasibility of deploying deep learning models for pest identification in agricultural settings and providing a scalable framework adaptable to various crop systems. It advances existing literature by integrating drone-based image acquisition with edge computing for real-time analysis, thereby addressing limitations of prior studies that relied solely on stationary ground cameras or off-line processing. The study underscores the importance of multidisciplinary approaches combining crop science, computer vision, and human-centered design. The main conclusion emphasizes that the developed AI-based pest detection system can enhance precision agriculture practices, reduce dependency on manual scouting, and enable proactive pest control measures. It recommends expanding the dataset to include emerging pest species, integrating multi-spectral imaging for improved detection accuracy, and developing user-friendly mobile applications to facilitate widespread adoption among smallholder and commercial farmers. Future research should explore longitudinal impact assessments and the integration of predictive modeling for pest outbreak forecasting, further optimizing crop protection strategies through technological innovation.