This work aimed to predict mass (weight) and shape ratio (format) of yellow melon through computer vision techniques (VC). To do this, a digital camera was used to take pictures of all melons (n=135). The images processing consisted in filtering colors in the RGB space, thresholding by Otsu's method and, finally, detection of melon's contours. The used processing techniques were sufficient to
... [Show full abstract] separate the melon from the image background, allowing calculating the area of the melon (Amelon), in both square pixel (pixel²) or square centimeters (cm²), which showed very strong Pearson’s correlation (0.993**). By using area-based linear regressions, it was possible to predict the weight of the melon in kilograms, from Amelon in pixel² (Pearson’s correlation = 0.993**) or cm² (0.989**). The shape ratio (SR) has estimated based on melon’s diameters (L - longitudinal and T - transversal), which were obtained using pachymeter (real) or computer vision (CV). Based on real data set (pachymeter), melons were classified by SR into four groups, considered as reference. The based-CV algorithm was able to classify the same melons in the same groups with hit percentage of 96%. The correlation between the product of the multiplication of both diameters (Lcv*Tcv) and the melon area (Amelon, pixel²) was very strong (0.9987**), and the coefficient of calibration of 93 pixels per centimeter presented a good fit. Based on these two results, the weight, in kilograms, and both diameters, in centimeters, were predicted from the measurements of L and T, in pixel, obtained by computer vision.