TECHNOLOGIES FOR QUALITY MEASUREMENT AND GRADING OF FRUITS AND VEGETABLES
Location: Sugarbeet and Bean Research
Title: Analysis of hyperspectral scattering images using locally linear embedding algorithm for apple mealiness classification
| Huang, Min - |
| Wang, Bojin - |
| Zhu, Qibing - |
Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: September 3, 2012
Publication Date: November 1, 2012
Citation: Huang, M., Wang, B., Zhu, Q., Lu, R. 2012. Analysis of hyperspectral scattering images using locally linear embedding algorithm for apple mealiness classification. Computers and Electronics in Agriculture. 89:175-181.
Interpretive Summary: Mealiness in apples is a symptom of physiological disorder, which is often related to the over-ripening of fruit at harvest and high temperature/humidity during postharvest storage. Mealy apples have an undesirable texture and taste, and thus are not suitable for the fresh market. A nondestructive technique for detection and segregation of mealy apples is needed to ensure the quality of fresh apples delivered to the marketplace. In this research, hyperspectral scattering technique was used for detecting the mealiness of 'Delicious' apples that had undergone two different postharvest storage treatments [i.e., refrigerated air at 4 degree C and high temperature (24 degree C)/high relative humidity (95%)]. The technique enables assessing the structural/physical properties of fruit through quantification of light scattering in the fruit for a spectral region. Destructive confined compression tests were carried out to determine the hardness and juiciness of each apple, which were used as a measure of apple mealiness. An image analysis method, referred to as the locally linear embedding (LLE) algorithm, was developed to extract hyperspectral scattering features for 580 'Delicious' apples for the spectral region of 600-1,000 nm. The mealiness classification model, developed based on the LLE algorithm, achieved 80.4% accuracy for normal and mealy apples, compared with 73.0% by the mean spectra method that was developed in a previous study. For apple samples with a greater degree of mealiness, the classification model was able to achiever a higher accuracy of 85.6%. This research demonstrated that hyperspectral scattering technique, coupled with the LLE algorithm, is potentially useful for detection of apple mealiness.
Hyperspectral scattering technique provides a means for assessing the structural and/or physical properties of apples. It could thus be useful for detection of apple mealiness, which is a symptom of physiological disorder, resulting in an undesirable texture and taste for apples and degrading their marketability. This research was aimed at developing and evaluating a locally linear embedding (LLE) algorithm, a features extraction technique, to extract spectral features from the hyperspectral scattering images for mealiness classification. Hyperspectral scattering images between 600 nm and 1,000 nm were acquired for 580 'Delicious' apple using a hyperspectral imaging system. Destructive confined compression tests were performed to determine the mealiness of apples in terms of hardness and juiciness. LLE algorithm was developed to reduce the dimensionality of, and extract features from, the hyperspectral scattering image data. Partial least squares discriminant analysis, coupled with leave-one-out cross validation, was applied to develop classification models based on the LLE and mean spectra algorithms. The model based on the LLE algorithm achieved an overall classification accuracy of 80.4%, compared with 73.0% by the mean spectra method for two-class classification (i.e., mealy and nonmealy) of all samples. For a subset of 290 samples with greater variations in mealiness, the LLE algorithm had an overall classification accuracy of 85.6%, compared with 82.5% by the mean spectra method. Hence, the LLE algorithm provided an effective means to extract hyperspectral scattering features for mealiness classification.