An Efficient Deep Feature Fusion-Based Convolutional Neural Network for Multi-Class Millet Classification
TL;DR
The proposed intelligent classification framework provides a scalable solution for automated grain inspection, food authentication, precision agriculture, and smart agricultural decision-support applications and achieves superior classification capability with high prediction reliability and strong generalization performance compared with conventional deep learning architectures.
Abstract
Accurate identification of millet varieties was essential for improving food quality assessment, agricultural productivity, post-harvest management, and intelligent grain processing systems. Conventional manual classification methods are often time-consuming, dependent on expert knowledge, and affected by visual similarities among different millet categories. This research proposes an efficient Deep Feature Fusion-Based Convolutional Neural Network (DFF-CNN) framework for automated multi-class millet classification using digital grain images. The developed approach integrates hierarchical feature extraction and multi-level feature fusion to combine low-level textural characteristics with high-level semantic representations, enabling improved discrimination among visually similar millet varieties. The image dataset consisting of multiple millet classes was subjected to preprocessing operations including resizing, normalization, noise reduction, and augmentation to enhance image quality and model robustness. The proposed architecture utilizes convolutional feature extraction, feature fusion, attention-based refinement, global average pooling, and Softmax-based classification to achieve accurate recognition. The performance of the developed model was evaluated using multiple evaluation parameters, including accuracy, precision, recall, F1-score, specificity, Matthews Correlation Coefficient, Cohen’s Kappa, confusion matrix, and ROC-AUC analysis. Experimental results demonstrate that the proposed framework achieves superior classification capability with high prediction reliability and strong generalization performance compared with conventional deep learning architectures. The generated feature representations effectively capture important grain characteristics such as texture patterns, morphological structures, and surface variations, resulting in reduced classification errors among millet categories. The proposed intelligent classification framework provides a scalable solution for automated grain inspection, food authentication, precision agriculture, and smart agricultural decision-support applications.