Contrasting bean analysis system based on YOLOv5 and a neural network model using the interval type-2 fuzzy set approach

Beans are a legume that has historically been a cheap source of protein, in the daily human diet. Therefore, it is challenging that the quality of this legume be guaranteed. In this work, we propose a novelty system for evaluating the contrasting bean quality, combining an algorithm for detecting black variety bean kernels and a neural network model based on an interval type-2 fuzzy set approach for weight estimation. To perform the detection task, YOLOv5 was used, which achieved favorable results, with an accuracy of 0.998, recall of 0.997, mAP_0.5 of 0.995 and mAP_0.5:0.95 of 0.733. Meanwhile, a model based on artificial neural networks was developed to estimate the weight. However, since the current quality assessment system includes vagueness and uncertainty, to deal with this problem, the interval type-2 fuzzy set approach was used by converting the collected data to fuzzy numbers with trapezoidal membership to use them in the training and validation process. The model for weight estimation showed favorable results, with a variance explained by the model (R²) of 0.99, a mean squared error (MSE) of 0.02, mean absolute percentage error of 0.05, and a mean absolute error (MAE) of 0.12 g. Meanwhile, an MSE of 0.02 and MAE of 0.11 g were the results for data not included in the training and validation process. Finally, both models were incorporated into a graphical user interface to facilitate their use in a real quality assessment process. The experiments with the proposed system demonstrated its effectiveness in detecting and estimating weight dynamically and with acceptable accuracy.