Crop pest classification using micro-doppler signatures via C-band radar in an anechoic chamber

• Development of a radar-based approach for early pest detection in agriculture. • Use of C-band Doppler radar to characterize spectral signatures of pest species. • Validation of multiple convolutional neural network (CNN) architectures for classification. • Achieved pest classification accuracy in validation between 91.7% and 97.1%. • Potential to enhance precision agriculture by enabling early pest control interventions. With the global population steadily rising, ensuring agricultural productivity requires more effective pest management strategies. Conventional detection methods, based on field scouting and visual inspection, often result in delayed responses and significant crop losses. This study introduces a novel approach that combines C-band Doppler radar with deep learning to detect and classify agricultural pests in cotton, citrus, and soybean crops. By capturing micro-Doppler spectral signatures generated by insect movements, radar provides distinctive features for species and life stage identification. Experiments were conducted in controlled conditions, where insects interacted with host plant leaves to simulate realistic field scenarios. Multiple convolutional neural network (CNN) architectures were evaluated, achieving validation accuracies between 91.68% and 97.1%, with a Macro AUC ≥ 0.992. These results confirm that radar signatures can reliably discriminate pest species and developmental stages. Beyond high accuracy, this method offers real-time, non-invasive monitoring, addressing limitations of traditional scouting techniques. The findings position radar sensing as a powerful tool for early pest detection and precision agriculture, enabling timely interventions that minimize yield losses. By integrating radar technology with advanced machine learning, this work contributes to more sustainable and resilient agricultural systems, supporting global food security through smarter pest management.