A Comparative Study of Machine Learning and Deep Learning Frameworks in Identifying and Detecting Nutrient Deficiencies in Plants using Leaf Image Analysis

Nutrient deficiencies in coffee leaves, including shortages of essential elements like nitrogen, phosphorus, and potassium, significantly impact crop yield and quality. These deficiencies manifest as visible symptoms in leaves, such as chlorosis, necrosis, and color variations, making image-based analysis a promising approach. This study compares the performance of deep learning and machine learning approaches for identifying nutritional deficiencies in coffee leaves through leaf image analysis. The study utilizes the co-leaf dataset, obtained from the Mendeley dataset repository, which contains labeled coffee leaf images exhibiting various nutrient deficiency symptoms. To improve reliability of classification, this approach integrates thorough data preprocessing and feature extraction techniques like color histograms and Local Binary Pattern (LBP). Conventional machine learning models including Random Forest and Support Vector Classifier (SVC) are utilised and achieved accuracy rates between 80% and 86%. In addition, the study evaluates deep learning architectures including ResNet, EfficientNetB0, InceptionV3, and VGG16, where VGG16 showed superior performance with accuracy rate of 92%. The results clearly indicate the advantage of deep learning techniques in diagnosing nutrient deficiencies, contributing to more efficient and accurate agricultural management practices.