Beyond pixels: deep feature extraction for plant nutrient deficiency classification
Anoop C. Patil, Ji-Yan Wu, Shalini Krishnamoorthi, Gajendra P. Singh, BongSoo Park, Daisuke Urano, Giovanni Volpe
Date: 24 August 2026
Time: 2:00 PM – 2:15 PM PDT
Place: Conv. Ctr. Room 2
Accurate detection of nutrient deficiencies from leaf images is critical for early nutrient stress identification and timely corrective intervention. We investigate controlled nutrient treatments to plant leaves representing complete absence (0%) and partial presence (5%) of iron, nitrogen, and phosphorus, alongside a healthy condition. The notation 0_X denotes complete absence of nutrient X (e.g., 0_Fe), while 5_X denotes 5% availability (e.g., 5_Fe) in plant leaves. We propose a two-stage framework in which a pretrained EfficientNet-B0 model extracts compact visual features from cropped leaf images,and lightweight machine-learning classifiers predict nutrient classes. Compared to raw RGB baselines,the deep-feature approach achieves higher classification accuracy, enabling robust identification of graded nutrient deficiencies for quantitative plant stress phenotyping.