Application of segmentation and edge detection algorithms during the microwave assisted drying of strawberries

VASCO, M. F.; GAMBOA SANTOS J.; CAMPAÑONE, L.A.

Conferencia; 34th EFFoST International Conference 2020; 2020

Image segmentation is a key step in most digital image analysis (DIA) involving computing tasks, as its performance directly affects the outcome of the image processing stages as a whole. In addition, Canny´s algorithm is considered one of the best contour detection methods using convolution masks based on the first derivative. In this way, a change in intensity manifests as an abrupt change in the first derivative, a characteristic that is used to detect an edge. The present work aimed to obtain relevant information from DIA on strawberries undergoing microwave drying (MW) treatments. The starting hypothesis is the feasibility of evaluating the changes in size, color and brightness that occur during the processing of strawberries in order to monitor the loss of product quality. In this sense, images of strawberries obtained in offline mode every 10 min of drying (MW lasts 100 min) were analyzed and compared with the original images (time 0), using different segmentation algorithms (global, adaptive binary, Otsu and color segmentation in Hue-Saturation-Value, HSV, coordinates) and filters (mean, median and Gaussian). Them, edge detection was evaluated using Canny´s algorithm. The code was programmed entirely in Python (Anaconda platform) using the OpenCV library. Also, the NumPy and Matplotlib libraries were used to perform the data analysis.The algorithms proposed were useful to obtain relevant information about shrinkage and color losses during MW drying, highlighting the efficiency of DIA techniques to evaluate visual changes of foods subjected to processing. During the segmentation task the need of an effective prior edge detection of the slices in the MW plate was revealed. Canny´s algorithm implementation could detect strawberry edges, however, optimization in taking photos (brightness control and increased background contrast) could lead to a more efficient detection of fruits during the drying process previous to their segmentation step.