


Volume 20 No 20 (2022)
Download PDF
Dynamic Target Constraint Spectral and Spatial Convolution Neural Network towards Soil Nutrient Classification and Prediction for Suitable high yield Crop Recommendation
S.Devidhanshrii1, Dr.R.Shanmugavadivu
Abstract
Soil nutrient analysis is significant process in the precision agriculture which assist farmer with
suitable and efficient crop suggestion and recommendation towards potential yield by the
cultivation throughout the growing seasons on basis of its soil fertility content. Further it prevents
the major losses on the crop growth and productivity due to lack of nutrients. Nitrogen, prosperous
and potassium are considered as primary nutrient of any soil type. In order to acquire the soil data
for processing, hyperspectral imaging is used as effective mode as it is capable of discriminating the
soil characteristics on basis of its spectral signatures. Hyperspectral imaging is a non –invasive
technique and fast monitoring technique for accessing both conventional nutrient content of the soil
and conventional nutrient content of the crop as hyperspectral images with large terrestrial data
containing spectral reflectance value obtained using sensors through spectroscopy. On acquired
hyperspectral images, computer vision technique is employed to carry out nutrient prediction and
classification task. In order to carry out to those machine learning algorithm and deep learning
model employed as conventional techniques. Despite of many advantageous such as better
characterization and exploitation of the land surfaces by combining rich spectral and spatial
information’s, it leads numerous challenges such as curse of dimensionality and high intraclass
variability. To mitigate the above mentioned challenges, a new deep learning framework termed as
Dynamic Target Constraint Spectral and Spatial Convolution Neural Network encompassing of endmember extraction, selection, classification and prediction processed on the various layer of the
network. Initiallyhyperspectral image preprocessing is carried out to sharpening and smoothening of
the spectral bands of the images. Preprocessed image is employed to sparseregression which is
considered as spectral unmixing technique to extract the spectral characteristics of the different
materials within the mixed pixels and recover the spectrum of the each pure spectral signature
called end-member extraction. These extracted end members are employed to feature selection
technique in order to build the representative spectral libraries and towards quantitatively selecting
the subset of the spectra. It is further applied to the deep learning architectureto map multiple soil
nutrient variables as feature map in the pooling and convolution layers. Further spectral distribution
map helps to model the soil fertility index with nitrogen, potassium and prosperous contents and it
classify the soil nutrients into classes on basis of the soil fertilityin activation using the spectral band
value of the soil properties and fully connected layers and finally it predicts the suitable of the crop
for cultivation in softmax layer. Experimental analysis of the proposed architecture has been
evaluated on the Landsat-8 dataset which access the proposed performance of the dynamic target
constrained convolution neural network framework on the available spectral indices of the soil
contents against the existing deep learning based approaches. Proposed framework produces the
99.8% of the accuracy on classifying the soil nutrientparameters into fertile classes with respect to
different spectral wavelength on the soil fertility index and it exhibits superior performance
compared with other existing learning classification approaches
Keywords
Soil Nutrient classification, Convolution Neural Network, Spectral Reflectance Value, Hyperspectral Image, Endmembers extraction, Spectral Unmixing, Crop Prediction, Nitrogen Value, Chlorophyll Value, Soil Reaction pH
Copyright
Copyright © Neuroquantology
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Articles published in the Neuroquantology are available under Creative Commons Attribution Non-Commercial No Derivatives Licence (CC BY-NC-ND 4.0). Authors retain copyright in their work and grant IJECSE right of first publication under CC BY-NC-ND 4.0. Users have the right to read, download, copy, distribute, print, search, or link to the full texts of articles in this journal, and to use them for any other lawful purpose.