Development of a Machine Learning Digital Image Models of Groundnut Pods for Intelligent Threshing Machine Application Using Convolution Neural Network

Abstract

Groundnut is an oilseed crop cultivated for food, oil extraction, and industrial applications. Manual methods such as vernier caliper, ruler to measure groundnut pod dimensions are time consuming, prone to errors, and unsuitable for large scale processing. These limitations contribute to inefficiencies in threshing, such as increased groundnut pod breakage, foreign material contamination, and poor machine performance. This research employed machine vision and augmentation techniques to address these challenges by automating groundnut identification, classification, and sizing. Principal Component Analysis is applied to analyze the shape, size, and orientation of groundnut pods, enabling faster, more accurate, and simultaneous measurement of multiple pods far superior to manual linear measurements. To enhance classification, ResNet 101, a deep Convolutional Neural Network, was employed. This model enabled segmentation, feature extraction, identification and classification of different groundnut varieties such as Ex-Dakar, Jarma, and Samnut26. The trained model achieved a high accuracy of 98.6% and demonstrated strong performance across other evaluation metrics. Such as Ex-Dakar achieved 99% precision, 98% recall, 98% F1-score, and 100% ROC-AUC, Jarma scored 98%, 99%, 98%, and 100% and Samnut26 recorded 99%, 99%, 99%, and 100% respectively. This study establishes an intelligent framework for groundnut geometric analysis and classification using machine learning. The integration of PCA and deep learning not only improves accuracy and reduces human error but also supports the development of a smart, efficient groundnut threshing system that addresses postharvest processing challenges in agriculture.

Keywords: Convolutional Neural Network, ResNet 101, Groundnut pods, Principal Component Analysis, Classification