| Food sciences |
|
|
|
|
| Non-destructive detection of internal quality of apple based on CT image |
| HUANG Taotao， SUN Teng， ZHANG Jingping* |
(College of Biosystem Engineering and Food Science, Zhejiang University, Hangzhou 310058, China
) |
|
|
|
Abstract China is a big producing country of fruits and vegetables, but the export price is quite low due to the lack of suitable grade classification technologies.
Domestic academics have been dedicating themselves to ameliorating the detection technology to change this situation. Xray computed tomography (CT), making use of the specific penetrativity, can acquire an accurate faulted image which contains many internal quality information of fruit.
In order to detect the character of apples quickly without damaging the sample simultaneously, a model was established based on the faulted image and some effective information related to the internal quality.
Although the average CT number of pulp area in an apple profile has a good linear relationship with the quality, it is of little practical value when popularized. We intend to make CT nondestructive detection method much more useful in the prediction of the apple quality. Firstly, the window/level number of CT image should be unified at an appropriate level, before building the model between CT numbers and gray level values. Secondly, the atactic pulp area should be separated from the faulted image by means of a segmentation algorithm named Otsu, which is an adaptive threshold method. Thirdly, the weighted mean of pixel numbers in this area should be calculated and converted into CT average numbers according to the relationship built before. Finally, the model of the relationship between the CT number and the internal quality in the area was developed. The model can be used to predict and analyze the apple quality. Generally, we had an arbitrary scale with air defined as having a CT number of -1000 HU and water of 0 HU. In our experiment, we detected the CT numbers of apple pulp ranging from -380 HU to 20 HU, which was will corresponded to the internal structure of apple. For the significant influence emerged by the window/level number in the process of converting the DICOM (digital imaging and communications in medicine) image into gray level image in BMP format, the window/level number must be unified. In comparison, we can get the clearest image at 430/-210. Then a regression model was set up between the CT numbers and the gray level values, which showed a good linear relationship with the R2 reaching 0.970 8. In consideration of the results obtained by some image segmentation algorithms, Otsu (maximum between clusters variance method) was put into use. It had different segmentation thresholds by computation, ranging from 71 to 91. At the same time, the weighted mean of pixel number can also be acquired from each gray level image by Otsu, and then the CT average numbers were converted from the pixel numbers. Finally the models between CT average numbers and the apple internal quality parameters were established, showing good linear relationships between CT number and the main under as sugar, titratable acidity and moisture, with the R2 values of 0.8464, 0.8233, 0.9075, respectively, and the prediction error can be controlled within 50%, 74%, 38%. It can be concluded that the internal quality of apple can be predicted by the CT faulted image quickly and nondestructively. We also found that the window/level number in the picture format of DICOM would significantly affect the gray level in BMP format, but in a fixed number, the two had a good linear relationship. The pulp area can be well separated from the whole image by Otsu, as well as figuring out the CT average numbers. At last we build three linear regression models between the number and the sugar, the titratable acidity, and the moisture separately, with good related coefficients and low forecast errors.
|
|
Published: 20 January 2013
|
基于CT图像的苹果内部品质无损检测
通过对苹果CT断层扫描图进行灰度分割，并在分离出其有效果肉区的基础上，成功地建立了基于果肉区CT均值的苹果整体内部品质的无损检测模型。首先统一CT图片的窗宽/窗位为430/-210，在此基础上确定CT值与灰度值的线性模型；然后利用Otsu自适应阈值法，对图像进行灰度分割，分离出苹果中心剖面的全部果肉区域，统计出该区域图像的灰度像素数加权均值，再转化成CT均值；最后建立上述中心剖面果肉区CT均值与苹果整体品质的关系模型，可以发现苹果的糖度、可滴定酸度、含水质量分数均与果肉区CT均值有较好的线性相关性，其R2值分别为0.8464、0.8233、0.9075，且平均预测误差分别小于5.0%、7.4%、3.8%。
|
Viewed |
|
|
|
Full text
|
|
|
|
|
Abstract
|
|
|
|
|
Cited |
|
|
|
|
| |
Shared |
|
|
|
|
| |
Discussed |
|
|
|
|
