Color grading methods of apple were discussed in this study by two types of neural networks used image processing systems instead of the human eye. The whole image data collecting system consisted of a turn table, a stage controller and a mono-axle driver was developed in this study. The results are summarized as follows:
1. Whole image data collecting system is useful to get a whole image of an apple, which is rolled out as one scene on a computer monitor like the perspective projections from the divided images. This system is proven to be useful in recognizing the size and the number of injured surfaces of apple.
2. Surface conditions can be classified into “normal red”, “injured colored red”, “poor colored red”, “vine” and “upper or lower background color” by the developed neural network. Judgement ratios classifying into these conditions, using this neural network under the three types of lamps, are more than 90%.
3. Another neural network was developed to grade the surface color of apple into “super excellent”, “excellent”, “good”, “poor colored” and “injured”. Grade judgement ratios for “super excellent”, “poor colored” and “injured” are very high, but for “excellent” and “good”, the ratios are not so high.
4. The special lamp, making the red color more clear, is helpful and useful to detect the injured apples easier.
View full abstract