Deep-Learning-Based Approach for Estimation of Fractional Abundance of Nitrogen in Soil From Hyperspectral Data

Abstract

One of the vital growth nutrient parameters of crops is soil nitrogen content. The ability to accurately grasp soil nutrient information is a prerequisite for scientific fertilization within the field of precision agriculture. Information pertaining to soil macronutrients, such as Nitrogen, may be obtained quickly through hyperspectral imaging techniques. Objective of this research is to explore the use of a deep learning network to estimate the abundance of urea fertilizer mixed soils for spectroradiometer data. The proposed approach was tested for silt clay and loamy types of soils. Spectral regions of 1899.2 nm for urea and 2195.1 nm for soils were identified as optimum spectral absorption features. The accuracy evaluation was performed using a linear regression model between actual and estimated abundances. At 1899.2 nm, the coefficient of determination (R 2) for mixed samples of urea and silt clay soil was found to be 0.945, whilst R 2 for urea mixed loamy soil were 0.954. Similarly, at 2195.1 nm, R 2 obtained 0.953 for urea mixed silt clay soil, and 0.944 for urea mixed loamy soil. The results show that the estimated abundances obtained through the derivative analysis for spectral unmixing (DASU)-based deep learning network, facilitated a greater accuracy in comparison to the sole use of DASU. These results were then verified through conventional chemical analysis methods. The outcome of this study determines the abundance of urea mixed soils. Therefore, it is inferred that the hyperspectral imaging technique may be utilized in-situ to assess the agricultural land's soil fertility status.

Residual Dense Autoencoder Network for Nonlinear Hyperspectral Unmixing

“…We define (4) where ⊙ is the Hadamard product; and denote the ith column of E and A, respectively; is the interaction coefficient between the ith and jth endmember in bilinear terms, describing the second reflection path as being much longer with a weaker signal intensity than the first. Meanwhile, (3) and (4) must satisfy the physical constraints of ANC and ASC in (5).…”

“…VER the past few decades, hyperspectral images have been widely applied in agriculture surveillance [1], [2], soil analysis [3], [4], and geological surveys [5], [6] because of their high spectral resolution. Under the influence of limited spatial resolution and the complexity of the cover type distribution, mixed pixels inevitably exist in hyperspectral images, making it difficult to extract ground objects.…”

Application of Hyperspectral Technology Combined With Bat Algorithm-AdaBoost Model in Field Soil Nutrient Prediction

“…The introduction of machine learning theory in this field has become a research hotspot. Support vector machines [43], manifold learning [44], firefly algorithms [45], neural networks [46], random forests [47], decision-making trees [48], and other machine learning methods are used in soil classification, content estimation, model optimization, automatic interpretation, and feature recognition. The conventional method matches spectral data or their variation with laboratory data one by one, which are inputted into a computer to obtain the information discovery model with which to calculate the soil composition content represented by the unknown spectrum [49].…”

Unsupervised Unmixing and Segmentation of Hyper Spectral Images Accounting for Soil Fertility

“…Thus, by combining the concepts of precision agriculture and hyper-spectral imaging, a new level of analysis for soil fertility may be approached. With optimized analysis it has immense potential in the Indian subcontinent to aid in implementing modern farming techniques to maximize yields and profits for farmers [2].…”