Development of an expert system based on wavelet transform and artificial neural networks for the ripe tomato harvesting robot

摘要

Robots can reduce production costs in large-scale greenhouses. For automating the harvest operation, robot should be able to recognize the fruit. This study was conducted with a view to develop a robotic harvesting system that performs recognition of ripe tomato. For this aim, several color images of tomato plants were taken in RGB color model. After analyzing data colors, detection of ripe tomato was done in two steps, first removing background of image and then recognizing ripe tomato from unripe tomato. Removing background was done based on threshold method using R-G equation. The identification of ripe tomato from unripe tomato was performed by integrating image processing and artificial neural network. Image processing was done based on wavelet technique. Energy and correlation were defined as two wavelet features and totally 90 wavelet features were extracted from each tomato. Artificial neural network was used for classification of ripe tomato from unripe tomato. A feed-forward neural network with two hidden layers was developed. The 63% of samples were used in the training stage of neural network, 12% of samples for validation, and testing of the network was performed with 25% of samples. The proposed algorithm could classify ripe tomato and unripe tomato with acceptable accuracies of 95.45% and 90% respectively.