Submitted to: Transactions of the ASAE
Publication Type: Peer reviewed journal
Publication Acceptance Date: 1/1/2005
Publication Date: 10/1/2005
Citation: Qin, J., Lu, R. 2005. Detection of pits in tart cherries by hyperspectral transmission imaging. Transactions of the ASAE. Volume 48(5):1963-1970. Interpretive Summary: Each year, approximately 95% of tart cherries produced in the U.S. are pitted and processed into canned, frozen, and dried fruit or made into juice. The presence of pits or pit fragments in pitted cherry products makes the food industry liable for financial losses and poses a potential hazard to the consumer. To ensure finished products free of pits or pit fragments, an effective pit detection technology is needed. This research was to investigate the feasibility of using hyperspectral transmission imaging to detect pits in tart cherries. Hyperspectral imaging acquires both spectral and spatial information from an object simultaneously, which allows for better detection of some subtle/minor features in the object than conventional imaging or spectroscopy techniques. A hyperspectral imaging system was used to acquire transmission images from both intact and pitted cherries. Computer algorithms with the artificial neural network were developed for pits detection. The pit detection accuracy ranged from 90.6% to 100.0% with the average rate of 98.4%. Fruit size and defect affected pit detection accuracy and their effect could be minimized through proper selection of cherry samples for training the neural network. This research demonstrated that hyperspectral transmission imaging is useful for detection of pits in tart cherries. The research laid a foundation for further development of a pit detection system for assuring pit free cherry products. Such detection system will help minimize financial liabilities for the food industry and ensure consumer acceptance and safety.
Technical Abstract: Automatic detection and rejection of unpitted tart cherries will help the food industry in minimizing the product liability and maximizing the end-user and consumer acceptance of pitted cherry products. The objective of this research was to investigate the potential of hyperspectral imaging for detecting the pit in tart cherries. A hyperspectral imaging system was used to acquire transmission images from individual cherry fruit for four orientations before and after pits were removed over the spectral region between 450 nm and 1,000 nm. Cherries of three size groups (small, intermediate, and large), each with two color classes (light red and dark red) for two harvest dates were used for determining the effect of fruit orientation, size, color, and harvest time on the pit detection accuracy. Additional cherries were studied for the effect of defect (i.e., bruise) on the pit detection. Computer algorithms were developed using the neural network (NN) method to classify the cherries with and without the pit. Two data inputs, i.e., single spectra and selected regions of interest (ROIs), were compared for pits detection. The spectral region between 690 nm and 850 nm was most appropriate for cherry pit detection. The NN with inputs of ROIs achieved higher pit detection rates ranging from 90.6% to 100.0%, with the average correct rate of 98.4%. Fruit size and defect affected pit detection but its effect could be minimized by proper selection of cherry samples for training the NN.