Prediction of sugar content in greenhouse muskmelon based on machine vision

The morphology of fruit not only characterizes the variety characteristic of muskmelon, but also has high correlations with fruit maturity and quality. The machine vision technology was used to evaluate the sugar content of muskmelon fruits qualitatively and quantitatively. Firstly, a fruit image collection system was designed and developed, 45 muskmelon samples were collected from three different growth stages. The values of image features were calculated with RGB color model, L*a*b* color model, Gray level co-occurrence matrix (GLCM), and then input to the Back-propagation (BP) neural network to predict the glucose, fructose, sucrose and total sugar content. In the quantitative analysis, prediction models for each sugar were established. The result showed that the correlation coefficient between measured and predicted total sugar content is the highest 0.888. In the qualitative analysis, muskmelon growth stages were predicted through different sugar content combination values. In the experiment, 30 muskmelon samples were used to establish models as training set and 15 samples were used as test set, the predicted growth stages were in full accord with those real ones. The results showed that the application of machine vision technology has good prospects in the prediction of muskmelon's internal qualities.