Online Hyperspectral Line-Scan Fluorescence Imaging for Safety Inspection of Apples

A recently developed fast hyperspectral line-scan imaging system integrated with a commercial apple-sorting machine was evaluated for rapid detection of animal faeces matter on apples online. Golden Delicious apples obtained from a local orchard were artificially contaminated with thin smear of cow faeces. For the online trial, hyperspectral fluorescence images of 30 contiguous spectral channels from 400 to 700 nm were acquired from samples moving at a processing sorting-line speed of three apples per second. Based on fluorescence ratio as a multispectral image fusion method, a 100% detection rate (118 out of 118 faeces treated apples) with no false positives (0 out of 120 apples, 60 wholesome and 60 apples with defects acquired prior to the faeces treatment) were achieved. INTRODUCTION To minimize food-related illness, safe production of foods for consumption is a public concern (Mead et al., 1999). Fruits and vegetables contaminated with animal faecal matter are recognized as a major culprit for pathogenic E. coli O157:H7 (Armstrong et al., 1996; Cody et al., 1999). Optical imaging or machine vision techniques enable rapid quality and safety inspection of fruits and vegetables. Non-destructive sensing methods including visible/near-infrared (Vis/NIR) and fluorescence imaging techniques have been evaluated for quality and safety inspection of agricultural commodities (Chao et al., 2006; Chen et al., 1998; Kim et al., 2001; Liu et al., 2005). In particular, the efficacy of fluorescence imaging for post-harvest food safety inspection has been demonstrated using apples empirically contaminated with a range of diluted animal faeces (Kim et al., 2005; Lefcourt et al., 2003; Vargas et al., 2005). Researchers at the Instrumentation and Sensing Laboratory, USDA, have recently developed a prototype hyperspectral line-scan imaging system capable of reflectance and fluorescence measurements to inspect apples online for quality and safety. In this paper, results obtained from a line-scan hyperspectral imaging system integrated with a commercial apple-sorting machine for detection of faecal contamination on apples, mainly based on the fluorescence method, are presented. Apples used in this investigation were artificially contaminated with cow faeces to demonstrate rapid online detection of animal faecal contamination on apples. MATERIALS AND METHODS Online Hyperspectral Imaging System A prototype, online hyperspectral line-scan imaging system was developed to inspect apples for animal faecal contamination. The hyperspectral line-scan inspection system integrated with a commercial-grade apple-sorting machine (FMC Corp, Philadelphia, PA, USA) is shown in Fig. 1. Line speed of the apple-sorting machine was adjusted to run at approximately three apples per second. Proc. XXVII IHC-S8 Role of Postharv. Technol. in Global. of Hort. Eds.-in-Chief: E.W. Hewett et al. Acta Hort. 768, ISHS 2008 386 The hyperspectral line-scan imaging system utilizes an electron-multiplying charge-coupled-device (EMCCD: PhotonMAX, Roper Scientific, Inc., Trenton, NJ, USA). It has 512×512 pixels and is thermoelectric cooled down to -70°C via a three-stage Peltier device. The imaging device is coupled with a 10 MHz (pixel-readout rate), 16-bit digitizer. An imaging spectrograph (ImSpector V10, Spectral Imaging Ltd., Oulu, Finland) and a C-mount lens (Rainbow CCTV S6X11, International Space Optics, S.A., Irvine, CA, USA) are attached to EMCCD, respectively. The Instantaneous field of view (IFOV) is limited to a thin line by the spectrograph aperture slit (50 μm). Through the slit, light from the scanned line is dispersed by a prism-grating-prism device and projected onto the EMCCD. Therefore, for each line-scan, a two-dimensional (spatial and spectral) image is created with spatial along the horizontal axis and spectral along the vertical axis of the EMCCD. Reflectance and fluorescence imaging required a passive light source, and each method used independent continuous wave (CW) light sources. The illumination sources are 150 w, quartz halogen lamps and a high intensity UV lamp (ML-3500, Spectronics Corp., Westbury, NY, USA) for reflectance and fluorescence, respectively. Interface software (WinView/32 version 2.5.19.0) provided by the EMCCD manufacture was used for imaging system control and data acquisition. To increase imaging speed, the original image size, 512×512 pixels, was reduced by 6×6 binning to 85×85 pixels. The 6×6 binning and the apple-sorting machine speed resulted in the spatial pixel resolution of approximately 2 mm. Furthermore, it should be noted that not all pixels in the spectral dimension were utilized; the dispersed light by the spectrograph did not span the full vertical width of the EMCCD. Thus, it further reduced the effective spectral dimension to 60 pixels (channels) spanning from approximately 400 to 1000 nm with approximately 10 nm channel interval. A detailed description of spectral calibration is omitted for brevity. With UV-A, most of biological materials exhibit fluorescence emission from 400 to 700 nm. Thus, fluorescence spectra are presented only in that spectral range. We developed image processing and analysis software on a MS Visual Basic (Version 6.0) platform operating in Windows. Using the downloaded hyperspectral image cube data, it allows automated visualization of individual apple images and detection of faeces-contaminated spots as the stream of hyperspectral image cube data are accessed. We are currently incorporating the system data acquisition function to the software to achieve real-time visualization/detection. A preliminary test suggested that we could process over 50 apples per second using a PC with 2 GHz processor. With the current imaging configuration (e.g., hyperspectral), one of the limiting factors for processing more than three apples per second is the data transfer rate (i.e., 10 MHz pixel readout rate). The system can be configured to acquire only several spectral channels (multispectral mode) that will markedly increase the data transfer rate per line-scan. We are also in the process of updating the EMCCD with a greater than 30 MHz pixel readout rate. Apples and Animal Faeces A total of 60 apples, ‘Golden Delicious’ destined for making unpasturerized apple cider were randomly selected from a batch of over 500 samples obtained from a local orchard located in Maryland. Apples used for making apple cider or juice can include those with black pox, sooty blotch, bruises, cuts, rots, insect bites, and physical damages. In addition, 60 wholesome apples (without visual defects) were used as control samples for this study. Fresh cow faeces from animals fed feedstuffs containing green roughage were collected from USDA farm facilities in Beltsville, MD. A thin cow faeces spot (approximately 2 cm in diameter) was artificially created on each apple (destined for making apple cider) by smearing the cow faeces on the apple using a spatula. A total of 59 apples were treated with the cow faeces; one of the apples was left out while applying faeces smear. Note that the cow faeces smears created transparent film-like coatings on apples and. Visually, were not easily discernable by human eye.