SoybeanNet: A Lightweight Neural Network for Soybean Pod Detection and Quantification

¹ Department of CS & IT, Kalinga University, Raipur, India
² Research Scholar, Department of CS & IT, Kalinga University, Raipur, India

^* Corresponding author: ku.ashunayak@kalingauniversity.ac.in

Abstract

Accurate detection and quantification of soybean pods are essential for enhancing crop yield predictions and optimizing agricultural management practices. In this research, we introduce SoybeanNet, a lightweight neural network specifically designed to detect and count soybean pods across diverse agricultural environments. Leveraging recent advances in deep learning, SoybeanNet addresses challenges such as variable lighting conditions, occlusions from overlapping leaves, and varied pod orientations in the field. With its streamlined architecture, the model is both precise and computationally efficient, making it suitable for deployment on resource-constrained platforms like drones and mobile devices. To ensure robustness in real-world scenarios, the training dataset was augmented with diverse soybean plant imagery, enhancing the model's generalizability. Experimental evaluations demonstrate that SoybeanNet achieves superior detection accuracy compared to traditional image processing techniques and other lightweight models, maintaining consistent performance across different growth stages and environmental settings. Field trials further confirmed its rapid and accurate pod count estimation, contributing to improved yield predictions and informed decision-making for farmers and agronomists. This study underscores the potential of lightweight neural networks in precision agriculture, offering a scalable solution with low power consumption for real-time applications. Future work will focus on extending SoybeanNet to detect and quantify other critical crop features and integrating it with broader agricultural monitoring systems to support sustainable farming and food security.