EfficientNet-Based Feature Extraction and Ensemble Learning for Tomato Quality Classification

Tomato quality and maturation detection play a critical role in ensuring the freshness and ripeness of tomatoes in the agricultural and food industries. Traditional methods for evaluating tomato quality are often labor-intensive, prone to human error, and inconsistent. This paper presents a robust machine learning-based system for automating the detection of tomato quality and maturation stages using image data. We leverage EfficientNetB0 as the backbone for feature extraction due to its state-of-the-art performance in image classification tasks with minimal computational cost. The extracted features are then classified using an ensemble of traditional machine learning classifiers, including Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Gradient Boosting Machine (GBM), integrated through a stacking ensemble approach. To improve generalization, cross-validation techniques and model fine-tuning are employed, and model explainability is achieved through Grad-CAM visualizations, offering insight into model decisions. The system is evaluated on a dataset of tomatoes categorized as fresh, rotten, mature, and immature. Our approach achieves high accuracy of 80% in classifying tomato quality and maturation, demonstrating the effectiveness of combining EfficientNetB0 for feature extraction with ensemble learning for classification. This study introduces a novel hybrid system for automating the detection of tomato quality and maturation stages using advanced image processing and machine learning techniques. EfficientNet-B0 serves as the backbone for feature extraction, enabling high accuracy with minimal computational resources. Extracted features are processed through a stacking ensemble comprising Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Gradient Boosting Machine (GBM) classifiers. The model leverages Grad-CAM visualizations to enhance interpretability. Comprehensive evaluations on a diverse tomato dataset demonstrate an 80% classification accuracy, establishing this system as a scalable, efficient solution for precision agriculture. This work highlights significant improvements over traditional methods, reducing manual labor while increasing consistency in quality assessments.