Hyperspectral Image-Based Leaf-Level Spatial and Spectral Feature Mining for Phosphorus Deficiency Symptom Differentiation in Corn Plants at Early Vegetative Stage

Authors: JIN, JIAN - Purdue University; ZHANG, JINNUO - Purdue University; SONG, ZHIHANG - Purdue University; AMPONG, KWAME - Purdue University; BELTRAME, ANITA - Purdue University; ZHAO, TIANZHANG - Purdue University; Penn, Chad

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/19/2025
Publication Date: N/A
Citation: N/A

Interpretive Summary: Phosphorus (P) is a necessary nutrient for growth of all plants, including important agronomic crops like corn. Corn shows visible P deficiency symptoms only when the deficiency is extreme, otherwise it simply produces less yield. This decrease in yield without a visible deficiency is known as "hidden hunger". Being able to detect P deficiency in corn plants with hidden hunger would provide the grower with the ability to correct the deficiency before it is too late. The objective of this study was to develop a method for detecting hidden P hunger in corn plants with hyper spectral scanning. Corn plants were grown at low, medium, and sufficient P levels. Leaves were scanned at various growth stages with a multi-spectral scanner. Data were analyzed to determine the ideal spectral features and wavelengths related to plants with less than sufficient P levels. Corn was harvested at maturity and grain yield was measured. Several wavelength combinations were identified that were highly correlated to the plant P levels. These results will be used to develop in-field scanners for detecting P deficiency in corn so that growers can correct potential deficiencies if they exist, thereby maximizing yield.

Technical Abstract: Phosphorus (P) is a vital macronutrient for building up essential plant biomolecules. The accurate identification of plant P deficiency symptoms can provide effective crop management guidance and increase crop growers’ profit. Hyperspectral imaging (HSI) offers a real-time, non-destructive avenue for assessing crop nutrient status via its rich information in both spatial and spectral domains. However, limitations in current HSI devices and analytical algorithms have led most studies to concentrate solely on the spectral features. This study proposed a novel feature mining algorithm combining numerous spatial and spectral features to differentiate P deficiency symptoms in corn plants at the V6 vegetative stage. The leaf-level hyperspectral images from three levels of P treatments and two leaf positions were collected using an in-house developed handheld proximal hyperspectral imager, LeafSpec. Spatial and spectral features that exhibited significant differences between P treatments were generated by integrating well-designed spatial partition principles with spectral index images. The correlation coefficient between the P content and elected combination features was used as the metric to screen important features for the identification of P deficiency symptoms. The spatial and spectral joint effects showed a superior ability over common spectral indices in discriminating P deficiency at both leaf positions, especially for differentiation of corn plants receiving medium and sufficient P treatments. Feature visualization heatmaps also provided insights into different levels of P deficiency symptoms on the leaves. This study demonstrated advantages of the newly developed feature mining algorithm in searching effective spatial and spectral features for differentiating P levels at corn plants’ early vegetative stage.