Abstract
Efficient nitrogen management is critical for optimizing cotton yield, ensuring plant health, and minimizing environmental impact. Traditional nitrogen monitoring methods, such as soil sampling and laboratory analysis, are time-consuming, costly, and impractical for large-scale farming. To address this challenge, this study uses a multimodal deep learning model that integrates image-based analysis and soil nitrate measurements to estimate leaf nitrogen content, enabling accurate, real-time nitrogen assessment for precision agriculture. A convolutional neural network (CNN) is employed to extract features from leaf images related to nitrogen status, while incorporating soil nitrate data enhances the accuracy of leaf nitrogen predictions by accounting for variations in soil nitrogen availability. A potentiometric sensor, developed in this study, enables precise electrochemical measurement of soil nitrate concentration, ensuring the model benefits from a continuous and reliable soil nitrogen assessment rather than relying on commercial sensors. By incorporating sensor-based soil nitrate measurements alongside image features, this approach provides a more comprehensive and accurate evaluation of leaf nitrogen status. The CNN model’s performance is evaluated using R², precision-recall curves, and Receiver Operating Characteristic - Area Under the Curve (ROC-AUC) scores, demonstrating a higher predictive accuracy (R² = 0.94) in the multimodal approach compared to the image-only model (R² = 0.90) for 650 samples. Additionally, a CNN-based binary classification model categorizes nitrogen levels in cotton leaves into low and high nitrogen levels , enabling targeted fertilizer application and early nitrogen deficiency detection. The results confirm that the multimodal approach significantly outperforms image-only predictions, reducing misclassification errors and enhancing nitrogen management efficiency. This research contributes to sustainable and data-driven precision agriculture, providing farmers with a scalable, non-invasive tool to optimize nitrogen application, reduce the cost of fertilizer application, and improve cotton yield and productivity.
Date of publication
8-2025
Document Type
Thesis
Language
english
Persistent identifier
http://hdl.handle.net/10950/4881
Committee members
Shawana Tabassum, Md Masud Rana, Chi Tian
Degree
Masters in Electrical Engineering
Recommended Citation
Solaiman, Mohammad, "A MULTIMODAL DEEP LEARNING METHOD COMBINING IMAGE ANALYSIS AND SOIL NITRATE SENSOR DATA FOR ACCURATE ESTIMATION OF LEAF NITROGEN IN COTTON PLANTS" (2025). Electrical Engineering Theses. Paper 69.
http://hdl.handle.net/10950/4881