Quality Inspection Of Orange Based On Multi-spectral Imaging

Citrus is the fruit which is the most widely cultivated in the southern China. The citrus industry is the pillar industry of the rural economy in the southern China. Currently, the study on citrus's post-harvest grading is very little in our country. The classification method which we have now is mainly manual and mechanical grading. The two methods have low efficiency and poor accuracy. Therefore, the citrus industry in our country is lack of commercialization and its add-value is low. It has great significance to study how to grade citrus by non-destructive testing technology, because it can improve the quality of the citrus, enhance the citrus add-value and increase income of the farmers. According to the high efficiency of the computer vision technology and the object absorption and reflectance characteristics in different spectral. We study the method of citrus'grading based on computer vision technology and multi-spectral techniques. The main contents are shown as follows:1. A new method of multi-spectral images capturing with machine vision system based on tunable multi-spectral light is introduced, and a multi-spectral LED light sources suitable for this study is designed.2. The spectral relations of orange's different characteristics are performed using the multi-spectral LED light sources. The results are:the edge is easier to extract by the multi-spectral image under the red band and yellow band. The orange surface defects characteristic is easier to extract by the multi-spectral image under the red band and yellow band. The orange internal defect characteristic is easier to extract by the multi-spectral image under infrared spectroscopy, but surface defects characteristic can't be extracted by infrared spectroscopy band. So we determine defects in orange by the single-band images under infrared spectral with the multi-spectral images under the red band and yellow band together. the green image and a yellow orange' image have significant differences in gray scale in the red band, so the color characteristic is extracted by the two images under the red band and infrared spectra band together.3. The image pre-processing algorithms is studied. Through anglicizing of existing image pre-processing algorithms and the characteristics of the orange, the pre-processing algorithms used is introduced. The threshold is achieved by Otsu algorithm, and the edge is achieved by Roberts algorithm.4. The detected method for size and shape is introduced. The size characteristic is achieved by projection area method, and the shape characteristic is achieved by perimeter area ratio. Recognition is up to 96%.5. The method based on multi-spectral technology is introduced. The defects characteristic is achieved by two images, one captured under red and yellow band, another captured under infrared band. Recognition is up to 97%. 6. Color models, pseudo-color theory and pseudo-color theory are studied. The detected method for color is introduced. The color is achieved by the H component of the HIS modle. Recognition is up to 97%.