-
系统工作流程如图 9所示。设备启动时,大功率LED灯打开同时散热风扇开始工作,FPGA开始初始化配置并将掩模寄存器初始值清零;FPGA初始化完成后开始初始化配置相机寄存器;相机初始化完成后系统开始采集图像信息,为确保图像输出稳定系统丢弃前10帧图像;当设备距离被测物体40 cm以内时, 系统启动图像处理功能,一方面, RGB图像与掩模寄存器中保存的掩模进行融合, 并将带有残留物标记的图像传送到RGB触控屏中,另一方面,RGB图像通过分块大津算法生成新的残留物掩模并更新掩模寄存器; 而当距离超过40 cm时,RGB图像直接显示在RGB触控屏;在RGB屏中可以选择是否将融合后的荧光图像进行保存。
-
当FPGA初始化完成后,FPGA需要配置相机的252个寄存器。相机配置状态机如图 10所示,状态机从idle跳转到wrreg_req状态并向IIC控制器发送写命令,IIC控制器开始向相机一个寄存器地址写入命令,写入完成后rw_done置1,状态机跳转为judge状态并判断相机寄存器配置完成个数,当计数器cnt小于252时,状态机跳转回wrreg_req状态继续写命令; 当cnt等于252时,状态机跳转到finish状态,从而完成相机的初始化配置。
相机部分配置参数如表 1所示。其中16进制数0 x 3808~0 x 380 b为图像分辨率寄存器地址范围,0 x 3503为手动曝光寄存器地址,0 x 3500~0 x 3502为曝光时长寄存器地址范围,0 x 4300为格式控制寄存器地址,0 x 501 f为图像输出格式选择寄存器地址。
表 1 相机主要配置参数
Table 1. Main camera configuration parameters
register name address value TIMING DVPHO H 0x3808 0x03 TIMING DVPHO L 0x3809 0x20 TIMING DVPVO L 0x380a 0x01 TIMING DVPVO L 0x380b 0xe0 AEC PK MANUAL 0x3503 0x03 AEC PK EXPOSURE 0x3500 0x00 AEC PK EXPOSURE 0x3501 0x6e AEC PK EXPOSURE 0x3502 0x30 FORMAT CONTROL 0x4300 0x61 ISP FORMAT 0x501f 0x01 -
相机接口逻辑主要完成相机RGB565格式输出数据的并行接收,其接收功能主要由10个不同位宽寄存器和2个数据选择器组成。如图 11所示,其中寄存器r_Data、r_Href、r_Vsync分别对输入数据Data、Href、Vsync、进行打一拍操作。H_count对每行像素数据进行计数,当H_count为偶数时将r_Data中8位图像数据保存在r_DataPixel高8位中;H_count为奇数时将r_Data中8位图像数据保存在r_DataPixel低8位中。在完成16位数据接收后,寄存器r_DataValid置1,表明此时r_DataPixel中16位图像数据为正确数据,可以输出到DataPixel口。为确保图像输出稳定,需要舍弃系统开始运行前10帧图像。寄存器FrameCnt对每帧图像进行计数,当FrameCnt计数到第10帧后dump_frame置1,表明此后图像有效。
-
触控屏接口逻辑是将RGB565格式图像数据流转换成VGA模式下的并行数据并发送到触控屏中。如图 12所示,根据触控屏接收时序要求,寄存器Hcount_r和Vcount_r分别产生行同步和场同步信号;Disp_Hs、Disp_Vs和Disp_De根据Hcount_r和Vcount_r产生数据有效信号;寄存器Disp_Red、Disp_Green、Disp_Bule分别将RGB565格式视频数据转换为并行数据。寄存器Frame_begin在场同步信号Disp_Vs上升沿置1,表明开始显示一帧图像。
-
当完成荧光图像的采集后,需要利用图像分割算法实现残留物的准确分割。由于具有计算简单、不受图像亮度和对比度影响的优点,大津算法在图像分割领域得到广泛引用。然而大津算法是一种全局阈值分割方法, 其在非均匀光照下会产生错误分割,并且当感兴趣目标灰度分布范围较大且部分目标灰度接近背景强度时,接近背景强度的部分目标会检测丢失。针对上述缺点,本文中基于分块大津算法的思想,通过对检测图像的合理分块,避免了背景子块误分割,提高了设备在不均匀光照或目标灰度差异较大情况下的检测能力。分块大津算法流程如图 13所示。首先FPGA从SDRAM缓存中读取荧光图像并经过灰度转换模块转换为灰度图像;然后FPGA按照行同步、场同步信号将一幅灰度图像划分M行N列的子块, 并运用大津算法计算每个子块分割阈值和类间平均灰度差;进一步对每个子块的类间平均灰度差进行判断,当子块类间平均灰度差较小时,该子块前景和背景灰度差异不大,判断为背景块,该子块分割阈值复位为0;当子块类间平均灰度差较大时,该子块中存在荧光残留物,保留计算的分割阈值。在系统实际使用中分块大津算法在设备距离被测表面40 cm以内启动,系统根据40 cm以内激发光源光强特点结合FPGA硬件资源占用将图像分为5行8列的子块进行处理,将判断子块是否为纯背景的类间平均灰度差设置为8。
-
分块大津算法在FPGA中实现如图 14所示。FPGA以数据流形式从SDRAM中读取的荧光图像Rgb_data缺少位置信息;为实现荧光图像准确分块,FPGA在统计模块中对经过的荧光图像数据流进行坐标统计并生成行同步信号Href、列同步信号Vsync和数据有效信号De;16位荧光图像数据经过灰度转换模块转换为8位灰度数据;系统通过调用8个大津算法模块和串行流水线设计方法实现分块大津算法计算。
分块大津算法中,使用大津算法模块计算每个子块的阈值和类间平均灰度差。如图 14中虚线框中所示,直方图统计模块具有5个输入,4个输出,其中输入Gray_data为8位图像灰度数据。输入Data_end在直方图统计完成后拉高,灰度阈值求完后拉低。输入Data_valid只在灰度直方图统计时为高,其余拉低。输入T为分割阈值, 输入Clr为清零信号,当灰度阈值求完后拉高使直方图统计模块清零复位等待下一次统计。在直方图统计模块中输出表达式[20]为:
$ S_0=\sum\limits_{i=0}^t n_i \\ $
(1) $ S_1=\sum\limits_{i=t+1}^{255} n_i $
(2) $ G_0=\sum\limits_{i=0}^t i n_i $
(3) $ G_1=\sum\limits_{i=t+1}^{255} i n_i $
(4) 式中,i为灰度值,t为灰度值,ni为灰度值等于i的像素个数,S0为灰度值不大于t的像素个数,S1为灰度值大于t的像素个数,G0为灰度值不大于t的像素灰度总和,G1为灰度值大于t的像素灰度总和。直方图统计模块输出经过后续乘法器、除法器、减法模块计算后得到类间方差,在比较器模块中保存类方间差最大时的阈值T,输出此时阈值以及类间灰度差。
手持式食品残留物荧光成像检测系统开发
Development of a handheld fluorescence imaging system for detecting food residues
-
摘要: 为了解决食品生产加工场所的卫生检查地点不固定、检测效果受环境因素影响大等问题, 根据不同种类食品残留物在紫外光照射下产生不同波长荧光的特性, 采用荧光成像技术和分块大津算法, 开发了一种能够协助现场检测员进行视觉卫生检查的手持式荧光成像设备。利用3种常见材料表面(木制案板、不锈钢板、聚乙烯塑料板)对奶粉(体积分数为50%、33%、20%、10%、2%)和菠菜残渣(体积分数为50%、33%、25%、20%、10%)进行了实验验证。结果表明, 该系统可以协助检测员进行卫生安全检查, 不同体积分数菠菜残渣在3种材料表面检出率为100%, 不同体积分数奶粉残留物在木制案板和不锈钢板表面检出率为100%, 在聚乙烯塑料板表面, 除体积分数为2%的奶粉残留物部分检出外, 其余体积分数奶粉残留物均能有效检出。该研究为食品加工场所手持式卫生检测设备开发提供了参考。Abstract: In order to solve the problems of food production and processing sites, such as the unfixed location of health inspection and the great influence of environmental factors on the inspection effect, according to the characteristics of different kinds of food residues under ultraviolet irradiation, a handheld fluorescence imaging device was developed by using fluorescence imaging technology and block Otsu algorithm, which can assist field inspectors in carrying out visual health inspection. The equipment detected milk powder (the volume fractions are 50%, 33%, 20%, 10%, and 2%) and spinach residue (the volume fractions are 50%, 33%, 25%, 20%, and 10%) on the surface of three common materials (wooden board, stainless steel board, polyethylene board). The results show that the system can assist inspectors in carrying out health and safety inspections. The detection rate of spinach residue with different volume fractions on the surface of the three materials is 100%, the detection rate of milk powder residue with different volume fractions on the surface of the wooden board and the stainless steel board is 100%, and on the surface of polyethylene plastic plate, except for the residual part of milk powder with the volume fractions of 2%, the remaining volume fraction of milk powder residue can be effectively detected. This study provides a reference for the development of handheld sanitary inspection equipment in food processing places.
-
表 1 相机主要配置参数
Table 1. Main camera configuration parameters
register name address value TIMING DVPHO H 0x3808 0x03 TIMING DVPHO L 0x3809 0x20 TIMING DVPVO L 0x380a 0x01 TIMING DVPVO L 0x380b 0xe0 AEC PK MANUAL 0x3503 0x03 AEC PK EXPOSURE 0x3500 0x00 AEC PK EXPOSURE 0x3501 0x6e AEC PK EXPOSURE 0x3502 0x30 FORMAT CONTROL 0x4300 0x61 ISP FORMAT 0x501f 0x01 -
[1] 李红秋, 郭云昌, 宋壮志, 等. 2019年中国大陆食源性疾病暴发监测资料分析[J]. 中国食品卫生杂志, 2021, 33(6): 650-656. LI H Q, GUO Y Ch, SONG Zh Zh, et al. Analysis of foodborne di-sease outbreaks in China in 2019[J]. Chinese Journal of Food Hygiene, 2021, 33(6): 650-656(in Chinese). [2] 李晓凤, 秦宇龙, 刘姜汝, 等. 即食果蔬中食源性致病菌风险评估的研究进展[J]. 食品与发酵工业, 2022, 48(18): 322-328. LI X F, QIN Y L, LIU J R, et al. Research progress in risk assessment of foodborne pathogens in ready-to-eat fruits and vegetables[J]. Food and Fermentation Industries, 2022, 48(18): 322-328 (in Ch-inese). [3] 黎远鹏, 黄富荣, 董佳, 等. 荧光光谱成像技术结合主成分分析与Fisher判别快速鉴别肉苁蓉[J]. 光谱学与光谱分析, 2015, 35(3): 689-694. LI Y P, HUANG F R, DONG J, et al. Rapid identification of cistanche via fluorescence spectrum imaging technology combined with principal components analysis and Fisher distinction[J]. Spectroscopy and Spectral Analysis, 2015, 35(3): 689-694(in Chinese). [4] 宋鑫澍, 陈国庆, 朱焯炜, 等. 3维荧光光谱测定白酒年份酒中乙酸的体积分数[J]. 激光技术, 2018, 42(4): 531-535. SONG X S, CHEN G Q, ZHU Zh W, et al. Determination of volume fraction of acetic acid in Chinese aged liquor by 3-D fluorescence spectrometry[J]. Laser Technology, 2018, 42(4): 531-535(in Ch-inese). [5] SHIBATA M, CHEN J, OKADA K, et al. Detection of food residues on stainless steel surfaces using fluorescence fingerprint[J]. Food Science and Technology Research, 2020, 26(3): 389-397. doi: 10.3136/fstr.26.389 [6] ABBAN S, JAKOBSEN M, JESPERSEN L. Assessment of interplay between UV wavelengths, material surfaces and food residues in open surface hygiene validation[J]. Journal of Food Science and Technology, 2014, 51(12): 3977-3983. doi: 10.1007/s13197-013-0927-9 [7] 康震, 丁利, 程云辉, 等. 基于上转换发光技术的便携式食品安全检测仪设计[J]. 传感器与微系统, 2019, 38(11): 114-116. KANG Zh, DING L, CHENG Y H, et al. Design of portable food safety detector based on up-conversion phosphor technology[J]. Transducer and Microsystem Technologies, 2019, 38(11): 114-116 (in Chinese). [8] 张利永. 基于智能手机黄曲霉菌毒素的快速检测[D]. 西安: 电子科技大学, 2019: 36-45. ZHANG L Y. Rapid detection of aflatoxin based on smartphone[D]. Xi'an: University of Electronic Science and Technology of China, 2019: 36-45(in Chinese). [9] SEO Y, LEE H, MO C, et al. Multispectral fluorescence imaging technique for on-line inspection of fecal residues on poultry carcasses[J]. Sensors, 2019, 19(16): 1-16. [10] BURFOOT D, TINKER D, THORN R, et al. Use of fluorescence imaging as a hygiene indicator for beef and lamb carcasses in UK slaughterhouses[J]. Biosystems Engineering, 2011, 109(3): 175-185. [11] GORJI H T, SHAHABI S M, SHARMA A, et al. Combining deep learning and fluorescence imaging to automatically identify fecal contamination on meat carcasses[J]. Scientific Reports, 2022, 12(1): 1-11. [12] SUEKER M, STROMSODT K, GORJI H T, et al. Handheld multispectral fluorescence imaging system to detect and disinfect surface contamination[J]. Sensors, 2021, 21(21): 1-15. [13] BECK E A, LEFCOURT A M, LO Y M, et al. Use of a portable fluorescence imaging device to facilitate cleaning of deli slicers[J]. Food Control, 2015, 51(5): 256-262. [14] WIEDERODER M S, LIU N T, LEFCOURT A M, et al. Use of a portable hyperspectral imaging system for monitoring the efficacy of sanitation procedures in produce processing plants[J]. Journal of Food Engineering, 2013, 117(2): 217-226. [15] LEE H, KIM M S, CHAO K, et al. Development of fluorescence based handheld imaging devices for food safety inspection[C]// International Society for Optical Engineering. Sensing for Agriculture and Food Quality and Safety Ⅴ. Bellingham, USA: International Society for Optical Engineering, 2013: 45-52. [16] OH M, LEE H, CHO H, et al. Detection of fecal contamination on beef meat surfaces using handheld fluorescence imaging device (HFID)[C]// International Society for Optical Engineering. Sensing for Agriculture and Food Quality and Safety Ⅷ. Bellingham, USA: International Society for Optical Engineering, 2016: 161-166. [17] EVERARD C D, KIM M S, LEE H. Assessment of a handheld fluorescence imaging device as an aid for detection of food residues on processing surfaces[J]. Food Control, 2016, 59(1): 243-249. [18] 孙颖馨. 基于FPGA红外成像光谱数据处理系统研究[J]. 激光技术, 2019, 43(6): 763-767. SUN Y X. Research of data processing systems for infrared imaging spectrometer based on FPGA[J]. Laser Technology, 2019, 43(6): 763-767(in Chinese). [19] 周小萌, 吴静静, 安伟. 一种鲁棒的非均匀光场中SD卡形态识别算法[J]. 传感技术学报, 2019, 32(4): 549-554. ZHOU X M, WU J J, AN W. A robust SD card shape recognition algorithm in inhomogeneous optical field[J]. Chinese Journal of Sensors and Actuators, 2019, 32(4): 549-554(in Chinese). [20] BARROS W K P, DIAS L A, FERNANDES M A C. Fully parallel implementation of otsu automatic image thresholding algorithm on FPGA[J]. Sensors, 2021, 21(12): 1-17. [21] EVERARD C D, KIM M S, LEE H. A comparison of hyperspectral reflectance and fluorescence imaging techniques for detection of contaminants on spinach leaves[J]. Journal of Food Engineering, 2014, 143(24): 139-145.