Diseño e implementación de un sistema de visión artificial para clasificación de al menos tres tipos de frutas

Abstract

This research presents artificial vision techniques applied to the detection of fruit’s features that can be used in the food industry. A computer vision system based in artificial neural networks and organized as a deep architecture comprises the main body of our method. The architecture is trained by using noise compensated learning, which creates a strong relationship between the artificial neural network and the object (fruit). This allows the recognition of complex features in selected fruits: strawberries, blackberries and uvillas; even with changing conditions in illumination, size and orientation. The programed algorithms use OpenCV libraries, which have been installed on Visual C++ Express 2010 free software. Also a database of fruits images captured through a high definition webcam was created.Edge detection algorithms and the use of a region of interest (ROI) allow the definition of a binary pixel image which constitutes the input for training the artificial neural network assigned to the tracking and tracing of the fruit. Isolated trainable agents take the information of the tracking network and use it for learning purposes. In this way important features concerning the shape and state of the fruits object of this investigation can be recognized. The systems have been tested in real time with real images. When a specific fruit kind is detected specific LED signals are activated.