Watermelons Talk: Predicting Ripeness through Tapping

Abstract

During the commercial production of watermelons, farmers must swiftly assess fruit ripeness post-harvest to minimize losses through sorting based on edibility time. This process enhances marketability and productivity but is often very tedious in traditional approaches. This article delves into the multifaceted realm of Internet of Things (IoT) based real-time watermelon ripeness evaluation. Watermelons, subject to diverse degrees of ripeness, significantly impact the fruit's taste and texture. Notably, watermelons cease to mature after detachment from the vine, underscoring the importance of selecting the ripest specimens at purchase. Prompt post-harvest fruit ripeness assessment is pivotal to mitigate losses, ensuring accurate sorting based on edibility timeline. Consequently, diligent watermelon ripeness assessment by farmers gains importance for enhanced marketability and productivity. While manual techniques like tapping, color examination, and day counting serve practical purposes, their accuracy relies on subjective judgment. Currently, the prevailing method for assessing watermelon ripeness is the sound test. This tapping technique surprisingly rests on logical grounds, as the resulting sounds offer an adequate ripeness indicator. However, personal interpretations of these sounds are influenced by subjective experiences and traditional wisdom. This article investigates non-destructive methodologies for evaluating watermelon ripeness. Then we propose WatermelonTalk, an IoT based real-time deep learning platform designed for acoustic watermelon testing. We also introduce the concept of the “tapping ensemble,” not previously found in the literature, which significantly enhances prediction accuracy. The article's contributions encompass the most comprehensive categorization of watermelons in the literature, specifically categorizing 1698 watermelons across 343 varieties by ripeness. Previous studies have considered either the 2-level test (unripe and ripe) or the 3-level test (unripe, ripe, and overripe). This article explores the 4-level test, where the unripe category from the 3-level test is further divided into the unripe class and the half-ripe class. In this test, the farmer pays more attention to the half-ripe class to ensure it undergoes more frequent testing than the unripe class. This precaution is taken to prevent these half-ripe watermelons from becoming overripe in the subsequent test. Our study achieved an enhanced testing accuracy of 97.64% for the three-level test and a notable accuracy of 94.07% for the four-level test, standing as the best result within the acoustic framework. The three-level test can be utilized by customers when purchasing watermelons, while the four-level test serves as a tool for farmers engaged in professional production.