Tomato ripeness and shelf-life prediction system using machine learning

This study proposes an ensemble approach to develop a tomato ripeness and shelf life prediction system based on defects and color intensity. The dataset has been created by designing an image acquisition system to capture 3450 images. The image outliers are overcome by various image processing steps. In the context of our proposed ensemble technique, we employ a variety of expert-based regressors, which encompass SVM, DT, RF, and GBM, in order to ensure distinctiveness. When assessing tomato characteristics for the purpose of quality determination, it becomes crucial to take into account attributes such as size, color, shape, texture, taste, nutritional content, defects, and ripeness, among others. A total of 13 features are manually extracted, whereas the Inception V3 model autonomously generates 2048 automated features. These 2048 features are then streamlined through the application of PCA dimensionality reduction, resulting in a final selection of 50 automated features. Altogether, the proposed work leverages a total of 63 features. The output of the ripeness regression models is divided into three classes based on their ripeness index and color magnitude to get the shelf-life of the tomatoes as Store, Sell, and Discount. The stacking technique is employed to achieve a final prediction of tomato shelf life with an impressive accuracy rate of 90.35%. These findings highlight that incorporating a variety of features, diverse pre-processing techniques, and proficient machine learning regressors can introduce substantial diversity in the ensemble approach, leading to enhanced accuracy when compared to conventional machine learning models. The proposed model has been rigorously compared with numerous state-of-the-art detection methods, yielding highly promising results.