Application of neural image analysis in evaluating the quality of greenhouse tomatoes

The paper presents the research on the use of methods of computer image analysis and artificial neural modeling in the process of assessing the quality of greenhouse tomatoes variety Cappricia. The subject of the study was tomatoes of the sizes from 40mm to 67mm and the colours: 1-6, include intermediate colours. The process of image acquisition and obtaining empirical data was conducted throughout the entire growing season in the period from the first harvest in the middle of May to the last harvest at the beginning of November. Satisfactory quality characteristics were obtained in the case of the RBF 37:37-39-1:1 and RBF 22:22-20-2:2 models. RBF 37:37-39-1:1 network, whose output variable was the colour of the tomato, the training quality was 0.930827, the validation quality was 0.911982, and the test quality was 0.979390. The RMSE rate of network training for the training set was 0.075986, for the validation set it was 0.072194, and for the test set 0.061714. For the RBF 22:22-20-2:2 network, the variables were colour and hardness. This network is characterised by a training quality of 0.985038, a validation quality of 0.990694 and a test quality of 0.985130. The RMSE rate of the network training is 0.065667, of the network validation it is 0.066187 and of the network test 0.073868. The research showed that performing a correct classification requires taking two digital images of the examined tomatoes, one of the stem and one of the front of the tomato, and generating training sets that contain average values for the extracted characteristics. (C) 2017 Elsevier B.V. All rights reserved.