基于图像传感技术的娃娃菜外观品质检测

目的设计一套基于图像传感技术分析娃娃菜外观品质如尺寸、重量、瑕疵点等的检测方法。方法搭建一套图像采集平台拍摄娃娃菜不同侧面,应用图像处理技术,分割出娃娃菜图像区域,并数字化其区域特征信息(包括:投影面积、尺寸、瑕疵点面积等)。结果建立合格娃娃菜的侧面投影面积与重量真实值间的线性关系,其相关系数为0.938,均方根误差为36.52 g;对比人工检测,图像法可以识别出娃娃菜外表面95%的瑕疵点(腐黑点、裂纹裂缝等);参照娃娃菜的分级标准,以图像法获取各特征指标,结合聚类算法分级娃娃菜,其中K-medoid法准确率为100%,Gath-Geva法准确率为96.67%。结论机器视觉技术可应用于娃娃菜的自动检测和分级,为在线无损检测提供参考。     

基于机器视觉的苹果表损智能检测系统设计

[目的]满足苹果外观品质和大小综合分级的现实需求,解决中国苹果人工分选效率低,分选设备结构复杂、成本高等问题。[方法]提出一种YOLOv5s-apple模型,在主干网络中引入Transformer模块和CBAM注意力模块,同时加入加权双向特征金字塔网络（Bi-FPN）改进颈部网络,并结合HALCON软件,利用自行设计的一种苹果表损智能检测系统进行表损分拣和大小分级。[结果]与原YOLOv5s模型相比,YOLOv5s-apple模型的mAP提升了6.2%,检测系统的分拣分级准确率可达97.5%,试验系统的处理速度为5 s/个。[结论]试验系统可以有效地进行苹果分级分选。     

On-line detection of eggshell crack based on acoustic resonance analysis

An on-line system based on acoustic resonance was developed for eggshell crack detection. The system employed digital signal processing (DSP) as core processor to collect and analyze the response signal of shell eggs. The effects of excitation point, speed of the conveying system, impact intensity and egg mass on the frequencies response signals were investigated. Based on previous experiments, differences between the signals of intact eggs and cracked eggs were analyzed. Five excited resonance frequency characteristics expressing the differences between intact and cracked eggs were explored as input vectors. A simple and robust calibration model was built to discriminate intact and cracked eggs, while meeting speed requirement of on-line measurement. The identification rates for intact and cracked eggs were 100% and 96.1% respectively. It is concluded that the system has significant potential in the on-line detection of cracked eggs.     

基于Welch法功率谱和广义回归神经网络的禽蛋裂纹辨识

为建立一种快速无损检测禽蛋裂纹的方法,构建了基于磁致伸缩振子扫频式振动的禽蛋裂纹检测系统。系统以声学特性为基础,通过利用Welch法功率谱分析禽蛋振动音频信号,利用主成分分析法提取特征向量中的有用信息并构建基于广义回归神经网络（generalized regression neural network,GRNN）的禽蛋裂纹检测模型。实验对290 枚鸡蛋进行检测（训练集200 枚,测试集90 枚）。结果表明,测试集中无损蛋与裂纹蛋的判别率分别达到96.7%和98.3%。研究表明,利用磁致伸缩振子扫频和Welch法功率谱分析,通过主成分分析法提取特征向量中的有用信息并结合GRNN模型检测禽蛋裂纹是可行的。     

Motion compensated image processing and optimal parameters for egg crack detection using modified pressure

Shell eggs with microcracks are often undetected during egg grading processes. In the past, a modified pressure imaging system was developed to detect eggs with microcracks without adversely affecting the quality of normal intact eggs. The basic idea of the modified pressure imaging system was to apply a short burst of vacuum within a transparent chamber in order to cause a momentary and forced opening in the egg shell with a crack and thus to utilize the changes in image intensities during this process. The intensity changes from dark to bright in the shell surface were recorded by a high-resolution digital camera and processed by an image ratio technique. However, the performance of the imaging system was compromised by both false readings due to motion of intact eggs relative to the camera and an improper selection of parameter values for the detection algorithm. First, a machine vision technique based on motion estimation of individual eggs was developed to compensate any motion errors present on images and thus reduce false crack-detection readings. The simulation results of the developed motion estimation and compensation technique with 3,000 eggs showed no false errors. Second, the receiver operating characteristic (ROC) curve was used to evaluate and compare the performance of the crack-detection algorithm under varying parameters (ratio and detection-tolerance thresholds) and to find the optimal parameter values. The area under the ROC curve (AUC) was used to compare the performance under varying parameter values. The minimum distance and Youden index criteria were used to find the optimal values from the ROC curve. The minimum distance criterion found the optimal parameters at 1.11 and 20 (or 1.1 and 25) for the ratio and detection-tolerance thresholds, respectively. The true positive and false positive rates at the optimal conditions were 98.91 and 0.14 %, respectively.     

基于改进YOLOX模型的樱桃缺陷及分级检测

目的:实现工业化条件下樱桃的快速分级。方法:采用YOLOX网络对缺陷果进行检测,通过为特征金字塔网络设置适当的融合因子来提高不明显缺陷的检测精度,并将Focal Loss集成到损失函数中;使用YOLOX网络对完好果进行分级,引入注意力机制CBAM来加强网络特征提取。结果:樱桃表面缺陷的平均检测精度为97.59%,大小和颜色分级的平均检测精度为95.92%。结论:改进后的YOLOX网络可明显提升樱桃缺陷及分级检测的精度。     

禽蛋品质无损检测及分级技术研究进展

禽蛋产业是世界农业产业的重要组成部分,禽蛋产业也是我国农业的关键组成,是关系民生和社会稳定的重要产业。目前,禽蛋企业主要是依靠人工挑选及分级,劳动强度大且效率低,检测精度无法保证。因此,禽蛋品质的快速无损检测及分级研究对产业发展具有重要意义。本文根据现有研究,针对禽蛋无损检测中广泛使用的声学检测、机器视觉检测、光学检测等方法的研究现状进行综述,阐明不同的无损检测方法在实际的禽蛋挑选及分级中的适用性。其中,声学检测主要应用于禽蛋表面的裂纹检测与分级;机器视觉技术主要应用于禽蛋的大小、颜色等外观参数检测;光学特性检测主要是用于禽蛋新鲜度等内部品质检测。本文有望为禽蛋产业智能化分级技术的发展提供一定的理论支持。     

基于机器视觉的轻量化芒果果面缺陷检测

目的：降低自动化芒果分级检测设备的制造成本。方法：对比了3种常用的目标检测算法在芒果缺陷检测中的效果,并以轻量化、移动设备可移植性为目标,提出了一种基于YOLOv5的轻量化芒果果面缺陷检测算法。结果：试验算法在满足芒果表面缺陷检测要求的前提下,相较于原算法可使参数量减少45.9%,浮点运算次数减少46.7%,权重文件大小减小45.2%。结论：试验算法有效降低了对部署设备的性能需求,在降低芒果分级检测设备的制作成本方面具有潜在价值。     

基于高光谱技术检测苹果外观缺陷

目的 利用高光谱技术检测苹果外观缺陷, 分析主成分分析法和波段比率算法研究高光谱图像的可行性。方法 在400~1100 nm波长范围内获取苹果表面的高光谱图像信息, 用主成分分析法处理高光谱下采集的苹果图像, 选取第三主成分图像进行分析, 作为最后的判别依据。波段比率算法中选取了717 nm和530 nm两个有效波段,将两个波段的图像进行比值运算。717 nm波段的图像进行阈值运算、中值滤波及形态学分析得到二值化掩膜图像, 再与二值化后的比率图像进行布尔运算, 提取缺陷的有效信息。结果 基于主成分分析法, 检测苹果表面缺陷的分级准确率为81.25%, 波段比率算法对苹果表面缺陷的分级准确率为93.75%。结论 利用高光谱成像技术下波段比率算法相对于主成分分成法更适合于实时、在线、快速检测。     

基于机器视觉的咖啡生豆快速检测与分级研究

目的 为了实现咖啡生豆的快速无损等级评估,建立一种基于机器视觉的咖啡生豆快速检测分级方法。方法 本次研究提取咖啡生豆图像预处理后的形状和颜色共13种特征,首先通过特征分布图分析与主成分分析找到13类缺陷豆的显著特征,并确定其显著特征值范围来判定缺陷豆,然后对正常豆按粒度大小判断其等级,最后在MATLAB App Designer平台上设计了咖啡生豆快速检测与分级界面。结果 研究表明,利用机器视觉技术能很好地识别各个类别咖啡生豆,检测一级咖啡生豆准确率为94.77%、二级为93.2%、三级为95.85%,13类缺陷豆平均准确率为82.25%,咖啡生豆检测平均准确率达到91.52%,检测300g咖啡生豆总用时25.3156s。结论 提高了咖啡生豆分级的智能化水平,分级过程平稳且快速,为今后咖啡生豆在线检测分选提供了技术支持。     

卷接机组故障诊断的模糊行为Petri网研究

根据卷接机组故障诊断推理中知识的模糊性,给出模糊行为Ｐｅｔｒｉ网（ＦＢＰＮ）的定义,研究用模糊行为Ｐｅｔｒｉ网表示模糊产生式规则的方法,提出一种模糊反向推理机制,给出算法的实现。最后以ＭＫ９—５卷接机组烟支空松故障为例,证明该方法的可行性和有效性。     

基于计算机视觉的茶叶品质在线评价系统

为实现茶叶品质在线评价及自动分级,摒除茶叶品质人工感官审评存在的缺陷,本文研发一套基于计算机视觉技术的茶叶品质在线评价及自动分级系统。采用Open CV和Visual C++开发茶叶品质在线评价软件,结合监督正交局部保持投影方法（supervised orthogonal locality preserving projections,SOLPP）对图像特征变量进行降维处理,对比随机森林（random forest,RF）、反向传播神经网络（back-propagation artificial neural network,BP-ANN）和相关向量机（relevance vector machine,RVM）茶叶品质在线评价模型,得出随机森林算法所建模型性能最好。系统自动完成茶样图像采集、原始图像预处理、特征提取、基于所建模型对待检茶样进行等级评价。控制系统根据评价结果,驱动分级及收集装置将检测茶样输送到相应等级槽中。经测试,研发系统对市售婺源仙芝绿茶、碧螺春绿茶的分级准确率达到93.00%以上。本系统结构简单,运行稳定,能将待检茶样准确送入到相应等级槽中,满足茶叶等级在线评价要求。     

基于SOPC水果分级检测系统研究

为了解决传统的水果分级检测系统存在的分级速度慢、系统庞大和成本高等问题,提出了基于SOPC水果分级检测系统,并将整个控制系统集成到一个FPGA芯片上,以脱离PC机平台,从而降低成本。通过Soble算法实现了水果图像的边缘提取,通过直方图的方法对水果的大小进行分辨,从而可以准确地对水果进行分级处理。测试结果表明,该系统具有检测速度快,开发周期短,可在线升级的优点,具有很好的应用空间。     

基于机器视觉的猴头菇品质快速无损检测与分级

传统的猴头菇品质检测与分级主要依靠人工分拣来完成,其主观性强、精度相差大、效率低,浪费了大量人力物力资源。鉴于此,为了实现猴头菇的快速无损等级评估,该研究引入机器视觉技术,提出了一种猴头菇品质的快速无损检测与分级方法,设计一套基于机器视觉的猴头菇品质快速无损检测与智能分级设备,并通过图像处理和软件设计开发一套猴头菇智能快速无损检测分级系统。通过加色法混色模型（RGB）对猴头菇的颜色特征的快速检测与等级的判定;采用图像阈值分割和Canny边缘检测,实现猴头菇完整度的判定;使用最小外接圆法对猴头菇的大小进行实时计算,完成猴头菇直径大小的判别;基于Microsoft Visual Studio 2017平台开发一套猴头菇品质快速无损检测可视化平台。试验证明,基于机器视觉的猴头菇品质快速无损检测与分级系统检测准确率达到97.07%,速度达到人工的5倍多。验证了系统的可靠性和可行性,为食品工业的智能化生产和加工提供了技术支撑,推动了机器视觉技术在食品行业的应用。     

基于改进CNN和数据扩充的苹果表面缺陷检测

目的:提高苹果表面缺陷的检测准确率和效率。方法:基于改进卷积神经网络(CNN)和数据扩充建立苹果表面缺陷检测方法。改建CNN的拓扑结构,并将其用于苹果表面缺陷检测;利用条件生成对抗网络,合成表面无缺陷和有缺陷苹果图像,实现图像数据扩充和提高改进CNN的苹果表面缺陷的识别性能;通过模型剪枝,合理权衡苹果表面缺陷的检测准确率、检测时间及节能限制,以提高所提方法的实用性。结果:当改进CNN中的解释层选用2 048个解释性神经元时,平均检测准确率最高;条件生成对抗网络增强了苹果图像数据集的多样性;随着增强图像数在测试数据集中占比的增加,所提方法对苹果表面缺陷的检测准确率不断升高;当剪枝后的模型尺寸占原始模型尺寸的百分比从100%降至50%时,可以以6.96%的准确率损失将苹果表面缺陷的检测效率提升1倍。结论:试验方法有望在苹果生产和加工过程中实现自动化缺陷检测。     

模糊控制查询表的快速获取及在造纸污水过程控制中的应用

由于模糊控制查询表在模糊控制中有着非常重要的应用,而基于传统利用MATLAB提供的模糊推理系统(FIS)的GUI界面来设计模糊控制查询表耗时较长。本文使用MATLAB设计了一种模糊控制查询表的快速获取方法,并通过该方法与GUI界面设计法获得的模糊控制查询表的比较,说明了该方法的可行性。同时,通过仿真实验验证了该获取模糊控制查询表方法的有效性。最后,将这种快速制取模糊控制查询表的方法应用于造纸污水处理过程中溶解氧浓度的模糊控制设计中。     

Hyperspectral reflectance and fluorescence line-scan imaging for online defect and fecal contamination inspection of apples

We have recently developed a rapid online line-scan imaging system capable of both hyperspectral Vis/NIR reflectance and fluorescence in the Vis with UV-A excitation. The hyperspectral online line-scan system was integrated with a commercial apple-sorting machine and evaluated to inspect apples for fecal contamination and defects at a processing line speed of over three apples per second. Results showed that fluorescence imaging (using a two-band ratio) could achieve detection of fecal spots on artificially contaminated apples with a 100% detection rate and no false positives regardless of the presence of defects. A NIR two-band reflectance ratio coupled with a simple classification method based on the mean intensity and homogeneity of the ratio achieved a 99.5% apple defect classification accuracy with a false positive rate of only 2%. The presented NIR processing regime overcame the presence of stem/calyx on apples that typically has been a problematic source for false positives in the detection of defects. The most significant and important outcome of this investigation is a line-scan inspection system that can potentially provide the capability for current sorting mechanisms, such as by size and color, as well as additional sorting for quality and safety attributes of food products. This line-scan based online imaging system offers great potential as a value-added dynamic inspection system due to its capability for multi-tasking to meet a variety of inspection objectives. A multi-tasking inspection system that can meet current industry sorting needs with the added benefit of safety inspection without requiring significant modification of existing infrastructure or incurring significant costs may lead the apple industry to consider adopting voluntary measures to further enhance safe production and processing of fruits. Company and product names are used for clarity and do not imply any endorsement by U.S. Department of Agriculture to the exclusion of other comparable products.     

Reduction in operator supervision time using a modified malfunction detection system with fuzzy inference in an industrial-scale fermentation of alkaline cellulase

A malfunction detection system based on fuzzy inference, in which a fermentation process is categorized into three states (normal, intermediate and abnormal) according to the computation of malfunction degree, was modified to reduce the supervision tasks required for an industrial-scale production of alkaline cellulase. The time lag of the culture, tau, was determined from the difference in the cumulated amount of CO2 evolved between the actual and standard values. By employing tau, the deviations of the actual values of process data (pH of broth and the amount of antifoaming agent supplied) from the standard values could be compensated for in terms of a delayed or forward effect in fermentation time. In the detection tests conducted for a series of 200 runs using the modified method, two runs and a single run were categorized as belonging to the abnormal and intermediate states, respectively, and all of the other runs were recognized to be in a normal state. As a result, the total supervision time for operators was reduced by the modification.     

基于可见-近红外高光谱技术对鸡种蛋性别鉴定的研究

目的 为实现鸡种蛋胚胎性别的无损检测,提出了基于可见-近红外高光谱检测海兰褐鸡种蛋胚胎性别的方法。方法 通过分析种蛋0~14 d大头部位的400~1000 nm波段下的光谱,建立基于偏最小二乘判别分析（PLS-DA）和支持向量机（SVM）的种蛋性别判别模型,比较不同孵育天数下的模型判别率,优选出最佳的检测天数;通过分析四种不同的预处理算法,选出最佳的鸡种蛋胚胎高光谱预处理方法,最后构建基于全波段和特征波段光谱信息的判别模型,并对结果进行比较。结果 基于PLS-DA和SVM的模型在第9 d的预测集结果达到最高,分别为80%和82.5%。主成分分析（PCA）结果表明,雄雌种蛋光谱信息可以进行区分;变量标准化（SNV）为最佳预处理方法;全波段相对于连续投影算法（SPA）、竞争性自适应重加权算法（CARS）选择特征波长的模型更优,建模集、预测集准确率分别为90%和85%。结论 研究结果表明可见-近红外高光谱技术可以快速、较准确、无损检测海兰褐种蛋胚胎性别,该技术为褐壳种蛋胚胎性别鉴定实现在线检测提供了一定的理论基础。     

Rapid and High-Throughput Determination of Melamine in Milk Products and Eggs by Full Automatic On-line Polymer Monolith Microextraction Coupled to High-Performance Liquid Chromatography

A rapid and high-throughput method for monitoring melamine (MEL) in milk products and eggs was presented. This method was based on a full automatic platform that was composed of on-line polymer monolith microextraction and high-performance liquid chromatography with ultraviolet detection. A poly(2-acrylamido-2-methyl-1-propanesulfonic acid-co-ethylene dimethacrylate) monolith was selected as the sorbent for purification and enrichment of MEL in milk products and eggs. A novel mixed-mode chromatographic column packed with octyl and sulfonic acid co-bonded silica was employed for quantitative determination of MEL in real samples. Several factors affecting extraction performance were investigated. Under the optimal conditions, the recoveries of MEL in milk products and eggs spiked at three levels of 0.5, 5.0, and 20.0 mg/Kg, ranged from 87.0% to 95.5%, with RSDs less than 6.4%. The limits of detection were 0.024 and 0.018 mg/Kg for MEL in the milk products and eggs, respectively.