Enhancing Lemon quality detection using Efficient-MobileNet with Explainable AI

Lemon, Citrus limon, is a bright yellow fruit known for its sharp, tangy flavor. With over 1,000 unique varieties worldwide, lemons hold significant economic value and are widely used in industries such as food and beverage, cosmetics, and cleaning. As an excellent source of vitamin C, lemons are highly valued for their quality, which directly impacts the economy, with the value of lemons for utilized production reaching approximately 698.3 million U.S. dollars in the 2023/2024 period. Ensuring top quality is essential; however, traditional detection methods are often costly, time-consuming, inconsistent, and prone to errors, making efficient quality detection crucial. To address these challenges, deep learning provides a standardized, efficient approach that is both quicker and more reliable. This study presents a highly optimized, lightweight CNN for intensive detection, integrating Efficient-MobileNet with the strengths of EfficientNetV2-S and MobileNetV2. By leveraging the design of MobileNetV2’s inverted residuals and linear bottlenecks, along with advanced training techniques and optimized layers from EfficientNetV2-S, the model delivers faster and more accurate performance while maintaining minimal parameters, making it ideal for resource-constrained environments. The dataset consists of 2,528 images of two different classes of lemon, along with an additional class that balances the dataset to prevent overfitting. The model’s rapid classification capabilities make it a valuable tool for improving lemon cultivation. Using the Efficient-MobileNet model, the generated accuracy reached 98.14%, with Precision, F1-score, and Recall all at 99.48%, respectively. These findings suggest that Efficient-Mobile is well-suited for accurately identifying lemon quality, which can enhance lemon cultivation and provide economic advantages to lemon farmers.