Notes

Abstract:
This paper presents a machine-learning approach for the detection of sweetness levels in watermelons. The model uses various physical and chemical attributes of the watermelons to predict the sweetness. Specifically, it uses the melons' skin image data to grade them into 3 levels of sweetness with A being the highest and C being the lowest. The model has been trained with data from 13 watermelons and uses a K-Means clustering algorithm to form clusters of the 3 grades. The model's accuracy has been determined as 84.62%, indicating a very high level of precision.


Intro:
Grading the quality of watermelons is a very time-consuming job and tends to be inefficient. The paper aims to solve it using machine learning, based on the physical as well as chemical attributes of the melons. It is important to have a good grading process so that consumer preferences are met. Reducing human intervention also increases the efficiency of the entire process, and results in better quality control. Also, traditional methods are often prone to errors due to the human sensor, and this makes the entire cycle from production to consumption less efficient and consumes more time.
We propose to develop a machine learning model for accurate prediction of sweetness level in a watermelon. The model will use a dataset of watermelon samples collected from various locations and their known sweetness levels by imagery as the training dataset. The input for the model will be the size, shape, and color, and the output will be each of the melons as a part of any of the clusters A, B, or C, which indicates the sweetness level.
The watermelons were graded into Grade A (High Sweetness Level), Grade B (Medium Sweetness Level), or Grade C (Low Sweetness Level). Most of the Yellowish-orange field spots and circle-shaped melons fell into cluster A, Yellowish-orange field spots and oval-shaped and Yellowish-white field spots and circle/oval-shaped melons fell into cluster B, and finally, perfect melons of any shape fell into cluster C.