Submitted to: Transactions of the ASAE
Publication Type: Peer reviewed journal
Publication Acceptance Date: 11/19/2002
Publication Date: 3/1/2003
Citation: LU, R. DETECTION OF BRUISES ON APPLES USING NEAR-INFRARED HYPERSPECTRAL IMAGING. TRANSACTIONS OF THE AMERICAN SOCIETY OF AGRICULTURAL ENGINEERS. v. 46(2). p. 523-530. Interpretive Summary: Bruises on apples are of great concern to the apple industry and the retailer because they lower the quality grade of fruit and can cause significant economic losses. It is a great challenge to develop a machine vision system to automatically detect bruises on apples because of the presence of fruit skin and because detection accuracies are affected by factors such as bruise type, age and severity, apple variety, and fruit pre- and post-harvest conditions. In this paper, we report on the development of a near-infrared hyperspectral imaging system and computer algorithms to detect bruises of different ages and severities. Results showed that the spectral region between 1000 nm and 1340 nm was most useful for bruise detection. Bruise features changed with time from lower reflectance to higher reflectance and the rate of this change was variety dependent. The system was able to detect both old and new bruises with the ecorrect classification rate up to 94% for Golden Delicious apples and up t 88% for Red Delicious apples. The optimal number of spectral bands needed for effective identification of bruises was between 20 and 40, with the corresponding spectral resolution between 8 nm and 17 nm. This research shows that near-infrared hyperspectral imaging offers great potential for effective detection of bruises on apples. The current system is suitable for offline inspection for defects on fruit. With improvement in image acquisition speed and detector technology, the technique can be adopted for online inspection of fruit for bruises and other surface defects. This would improve sorting efficiency and enhance fruit quality; thus, leading to savings in labor costs and the increased profitability of growing apples.
Technical Abstract: Development of an automated bruise detection system will help the industry and the retailer in providing superior fruit for the consumer and reduce potential economic losses. The objective of this research was to investigate the potential of near-infrared (NIR) hyperspectral imaging for detecting bruises on apples in the spectral region between 900 nm and 1700 nm. A NIR hyperspectral imaging system was developed and a computer algorithm was created to detect both new and old bruises on apples. Experiments were conducted to collect hyperspectral images from individual apple fruit over a period of 47 days after each fruit was bruised with different degrees of severity. Results showed that bruise features changed over time from lower reflectance to higher reflectance and the rate of the change was variety dependent. The spectral region between 1000 nm and 1340 nm was most appropriate for bruise detection. Using both principal component and minimum noise fraction transforms, the system was able to detect both new and old bruises, with the correct detection rate from 62% to 88% for Red Delicious apples and from 59% to 94% for Golden Delicious apples. The optimal number of spectral bands needed for bruise detection was between 20 and 40 with the corresponding spectral resolution between 8 and 17 nm. This research shows that NIR hyperspectral imaging offers great potential for effective detection of bruises and other surface defects on apples. With the improvement in image acquisition speed and detector technology, the NIR hyperspectral imaging technique can be used for on-line sorting of fruit for surface defects such as bruises.