Open Access

ARTICLE

Smart Nutrient Deficiency Prediction System for Groundnut Leaf

Janani Malaisamy^*, Jebakumar Rethnaraj

Department of Computer Science and Engineering, School of Computing, SRM Institute of Science and Technology, Chengalpattu, Tamil Nadu, 603203, India

* Corresponding Author: Janani Malaisamy. Email:

Intelligent Automation & Soft Computing 2023, 36(2), 1845-1862. https://doi.org/10.32604/iasc.2023.034280

Received 12 July 2022; Accepted 27 August 2022; Issue published 05 January 2023

Abstract

Prediction of the nutrient deficiency range and control of it through application of an appropriate amount of fertiliser at all growth stages is critical to achieving a qualitative and quantitative yield. Distributing fertiliser in optimum amounts will protect the environment’s condition and human health risks. Early identification also prevents the disease’s occurrence in groundnut crops. A convolutional neural network is a computer vision algorithm that can be replaced in the place of human experts and laboratory methods to predict groundnut crop nitrogen nutrient deficiency through image features. Since chlorophyll and nitrogen are proportionate to one another, the Smart Nutrient Deficiency Prediction System (SNDP) is proposed to detect and categorise the chlorophyll concentration range via which nitrogen concentration can be known. The model’s first part is to perform preprocessing using Groundnut Leaf Image Preprocessing (GLIP). Then, in the second part, feature extraction using a convolution process with Non-negative ReLU (CNNR) is done, and then, in the third part, the extracted features are flattened and given to the dense layer (DL) layer. Next, the Maximum Margin classifier (MMC) is deployed and takes the input from DL for the classification process to find CCR. The dataset used in this work has no visible symptoms of a deficiency with three categories: low level (LL), beginning stage of low level (BSLL), and appropriate level (AL). This model could help to predict nitrogen deficiency before perceivable symptoms. The performance of the implemented model is analysed and compared with ImageNet pre-trained models. The result shows that the CNNR-MMC model obtained the highest training and validation accuracy of 99% and 95%, respectively, compared to existing pre-trained models.

---

Keywords

---