Using Image Analysis to Predict the Weight of Alaskan Salmon of Different Species

ABSTRACT: After harvesting, salmon is sorted by species, size, and quality. This is generally manually done by operators. Automation would bring repeatability, objectivity, and record-keeping capabilities to these tasks. Machine vision (MV) and image analysis have been used in sorting many agricultural products. Four salmon species were tested: pink (Oncorhynchus gorbuscha), red (Oncorhynchus nerka), silver (Oncorhynchus kisutch), and chum (Oncorhynchus keta). A total of 60 whole fish from each species were first weighed, then placed in a light box to take their picture. Weight compared with view area as well as length and width correlations were developed. In addition the effect of “hump” development (see text) of pink salmon on this correlation was investigated. It was possible to predict the weight of a salmon by view area, regardless of species, and regardless of the development of a hump for pinks. Within pink salmon there was a small but insignificant difference between predictive equations for the weight of “regular” fish and “humpy” fish. Machine vision can accurately predict the weight of whole salmon for sorting.     

基于机器视觉的淡水鱼去鳞损伤无损检测识别方法

为了研究淡水鱼去鳞损伤无损检识别方法,以鲤鱼为研究对象,搭建视觉采集系统,实现图像采集与处理。通过采集鲤鱼去鳞后有去鳞损伤的图像,将样本图像划分为鱼腹、鱼背、鱼尾、损伤、背景五部分,提取感兴趣区域(ROI),比较其颜色差异,发现正常鱼体的表面区域与损伤区域颜色差异较大,以不同区域对应的红色通道值(R)、绿色通道值(G)、蓝色通道值(B)对应像素的平均值作为输入,建立基于广义神经网络(GRNN)和径向基函数神经网络(RBF)的鱼体损伤区域检测识别模型,经测试,识别准确率分别能达到98%和80%。通过计算预测损伤区域与实际损伤区域的相关面积大小,验证识别模型的准确度,经试验验证,基于GRNN和RBF网络识别模型对去鳞损伤预测的准确率分别达到了90%和83%。研究结果表明,机器视觉可应用于淡水鱼去鳞损伤的无损检测识别。     

巴美肉羊净肉质量与净肉率预测模型建立

在羊肉生产加工的屠宰线上,为了较快地预测胴体净肉质量和净肉率,并按照不同产肉性能对肉羊胴体进行等级划分,以建立科学合理的优质优价收购体系,本实验选取30头4～8月龄巴美肉羊,对其进行屠宰、冷却、分割,并分别测定其宰前羊体活质量（X1）、热胴体质量（X2）、眼肌面积（X3）、背肉厚度（X4）等指标。将数据采用SPSS 16.0软件进行相关分析和线性回归,最终得到以X1、X2和X3三因子共同预测净肉质量（Y1）的线性方程Y1=-1.113＋1.088X2-0.141X1-0.004X3（R2=0.996）和预测净肉率（Y2）的线性方程Y2=31.187＋2.362X2-1.090X1＋0.068X3（R2=0.920）,结果表明这两个方程均可在实际生产中用于巴美肉羊净肉质量和净肉率的预测。     

基于稻谷X射线荧光光谱测定快速识别糙米和精米中的镉含量

目的 基于稻谷镉的X射线荧光光谱测定, 建立糙米、精米镉的快速定量识别模型, 简化入仓稻谷重金属检测的砻谷、碾米等预处理步骤。方法 采用X射线荧光光谱法一一对应分析26组稻谷-糙米-精米样品中镉含量。根据线性、对数、逆、二次、三次、幂、指数等函数关系分别拟合稻谷-糙米、稻谷-精米、糙米-精米镉含量之间的回归模型。采用另外4组样品对决定系数(r2)大于0.95的模型准确性进行验证, 根据决定系数、误差值等筛选最优拟合模型。结果 稻谷-糙米-精米镉含量之间存在较强的相关性, 在此基础上建立的3个可食用米镉含量快速识别模型分别为: 稻谷-糙米三次函数回归模型Y=0.0131+0.7178X+0.5722X2-0.3492X3 (R2=0.9859); 稻谷-精米三次函数回归模型Y=0.0284+0.3779X+1.5500X2-1.2046X3 (r2=0.9855); 糙米-精米幂函数回归模型Y=0.9412×X1.0233 (r2=0.9902), 三个模型预测结果的绝对误差分别为8.91%、8.57%和10.24%。结论 本研究建立的回归模型具有良好的稻米镉含量相互预测性能, 该法有望简化稻谷镉检测前的砻谷、碾米等预处理流程, 提高检测效率。     

High-speed weight estimation of whole herring (Clupea harengus) using 3D machine vision

Abstract: Weight is an important parameter by which the price of whole herring (Clupea harengus) is determined. Current mechanical weight graders are capable of a high throughput but have a relatively low accuracy. For this reason, there is a need for a more accurate high‐speed weight estimation of whole herring. A 3‐dimensional (3D) machine vision system was developed for high‐speed weight estimation of whole herring. The system uses a 3D laser triangulation system above a conveyor belt moving at a speed of 1000 mm/s. Weight prediction models were developed for several feature sets, and a linear regression model using several 2‐dimensional (2D) and 3D features enabled more accurate weight estimation than using 3D volume only. Using the combined 2D and 3D features, the root mean square error of cross‐validation was 5.6 g, and the worst‐case prediction error, evaluated by cross‐validation, was ±14 g, for a sample (n = 179) of fresh whole herring. The proposed system has the potential to enable high‐speed and accurate weight estimation of whole herring in the processing plants. Practical Application: The 3D machine vision system presented in this article enables high‐speed and accurate weight estimation of whole herring, thus enabling an increase in profitability for the pelagic primary processors through a more accurate weight grading.     

A simple method for weight estimation of whole herring (Clupea harengus) using planar X-ray imaging

Fish weight is one of the key parameters used in commercial sorting of fresh fish. The capability of on-line and accurate estimation of the weight of each individual fish has a direct impact on profitability for the fish processors. Planar X-ray images of whole herring (n=16) were taken using a commercial medical X-ray apparatus. The integral intensity of all pixels in each image was calculated and correlated with the weight of the respective fish producing a linear relationship with a high correlation factor (R2=0.992) and a calculated mean prediction error of 2.5%. Planar X-ray imaging combined with simple image processing was successfully used to predict weight of whole herring. The technique has a potential to be used industrially for on-line weight estimation in sorting/grading systems.     

罗非鱼加工副产物不同部位硫酸软骨素的制备、理化性质及结构表征

采用双酶酶解、氯化十六烷基吡啶法从罗非鱼副产物不同部位（鱼头、鱼脊骨、鱼鳍、鱼尾）制备硫酸软骨素,利用紫外光谱、醋酸纤维电泳、红外光谱和高效液相色谱对其理化性质和结构特征进行分析。结果表明,鱼头、鱼尾所含硫酸软骨素主要为C型（Δdi6S）,含量分别为46.10%和41.01%;鱼脊骨、鱼鳍所含硫酸软骨素主要为A型（Δdi4S）,含量分别为71.86%和69.59%。纯度从高到低依次为鱼头90.37%、鱼尾83.33%、鱼脊骨59.76%、鱼鳍52.01%。鱼头、鱼脊骨、鱼鳍、鱼尾硫酸软骨素的单糖组成主要由不同比例的葡萄糖醛酸、葡萄糖和氨基半乳糖构成,mn依次为51 422、18 402、19 481、76 371。不同部位提取的硫酸软骨素的种类和组成有明显差异,其中罗非鱼头、鱼尾提取的硫酸软骨素与鲨鱼软骨来源硫酸软骨素结构相似,有望作为代替鲨鱼硫酸软骨素的潜在来源。     

基于多场耦合理论的稻谷储藏品质变化规律及预测模型的研究

为了探讨稻谷在储藏期间的品质变化规律并建立品质变化的预测模型,本文在多场耦合理论的基础上,利用自主设计的温湿度控制的模拟仓储存稻谷120 d,研究其不同储藏时间的水分含量、脂肪酸值以及过氧化氢酶活动度等品质指标的变化,根据模拟仓内温湿度耦合的规律与稻谷储藏品质的相关指标变化之间的关系,建立了稻谷储藏品质变化预测模型。结果表明：在稻谷储藏期间,温湿度均出现了耦合的现象,在粮堆中形成了温湿气的强弱耦合区域,在储藏第60d时,模拟仓内均出现了上层粮堆呈现一个“U”型高温高湿聚合区。60d后停止对冷热壁的制冷和加热温湿度的场强效应减弱,粮堆内没有无强弱耦合区的出现。温度和湿度的耦合作用对稻谷的水分含量的影响显著。稻谷储藏120 d时高温湿耦合区域稻谷的脂肪酸值达到接近重度不宜存的状态。而低温湿耦合区域和中间区域稻谷的脂肪酸值差异不显著。温湿度是影响稻谷过氧化氢酶活动度的主要因素,温湿度过高导致其活力下降。在温湿度耦合效应下稻谷脂肪酸值（Y1）和过氧化氢酶活动度（Y2）与温湿度和稻谷水分相关关系的模型分别为：Y1= -6.758 + 0.320X1- 0.081X2- 0.026X3 + 0.020X4 + 1.501X5,Y2 = 119.952-0.901X1 + 0.313X2 + 0.061 X3 - 0.044X4 - 4.088X5（X1：粮堆温度;X2：粮堆湿度;X3：粮堆深度;X4：储藏时间;X5：水分含量）。研究结果对于丰富多场耦合理论,指导智能化粮食储藏具有重要的意义。     

Application of computer vision for determining viscoelastic characteristics of date fruits

Date fruits are subjected to different environmental conditions from harvest to consumption. Since the rheological behavior of fruits are important for any processing operation, the viscoelastic characteristics of date fruits were studied at six moisture content, and four temperature levels of product, using a computer vision method. The whole date fruits were subjected to uniaxial compression for relaxing the stress during 300 s. As a new approach, a probe that equipped with a camera was used to measure the samples true contact area simultaneously. Three famous viscoelastic models were used to fit on experimental data. It was found that Generalized Maxwell and Nussinovitch models describe viscoelastic characteristics of date fruits appropriately, as compared to Peleg model. Furthermore, it was observed that moisture content of the product has dominant effect on viscoelastic characteristics of the fruits than temperature.     

基于机器视觉的单组分染液浓度快速检测方法

针对印染行业的单组分染液浓度的检测需求,提出一种基于机器视觉的对单组分染液浓度进行快速检测的方法。该方法采用颜色亮度可调的背光光源和彩色工业相机搭建机器视觉检测平台,将质量浓度范围在0~0.4 g/L的单组分染液样品通过玻璃皿置于背光光源上进行图像采集和处理,获取染液颜色特征值(R、G、B值),拟合不同光源条件下的R、G、B值随染液浓度变化的曲线,以朗伯比尔定律为依据建立R、G、B值与染液浓度的关系模型,实现染液浓度的预测。结果表明,对于DRA-3R染料溶液,光源为蓝光(亮度等级为50)条件下建立的最小值模型预测精度最高,平均相对误差绝对值为3.35%,并且该方法检测速度快、成本低,为染液浓度在线检测的工业化应用提供了一定的研究基础。     

基于机器视觉的香蕉果肉缺陷预测方法

针对香蕉内部果肉缺陷难以预测的问题,运用机器视觉技术对香蕉果皮与果肉进行图像识别,对识别参数进行数据拟合得到果肉缺陷的预测模型。将采集到的图像灰度化并进行滤波去噪,通过双阀值二值化和形态学分析对图像进行识别处理,提取香蕉果皮、香蕉果肉、香蕉果皮黑斑与香蕉果肉缺陷。计算提取区域的像素点总数,将其作为区域面积。分别用香蕉果皮总面积/香蕉果肉总面积与果皮黑斑面积/果肉缺陷面积之比来定义香蕉果皮黑斑度与果肉缺陷度。运用多项式拟合法,根据训练样本得出果肉缺陷预测函数,对预测函数进行残差分析。通过预测模型对香蕉划分等级,总准确率达到88.9%,与通过香蕉果皮进行等级划分其他方法相比,试验所得模型的预测准确率较高,表明通过香蕉果肉进行预测的方法具有一定的优越性。     

包装鲜鱼块形状对其品质的影响

以新鲜宰杀的草鱼为原料,采用普通包装的方式进行包装,并分别于8℃和20℃条件下贮藏,研究不同贮藏温度下,鲜鱼块的比表面积对其挥发性盐基氮(TVB-N)和硫代巴比妥酸值(TBA)的影响。结果表明,不同温度下,鲜鱼块的比表面积越大,TVB-N值和TBA值的增量随贮藏时间的增速越大。8℃和20℃贮藏条件下,包装鲜鱼块的TVB-N值的增量与贮藏时间及比表面积分别符合方程Y8=(0.00385+0.00000356)·S·t1.525和Y20=(0.00795+0.0000294·S)·t2.06。     

基于机器视觉的牡蛎分级设备设计

目的：提高牡蛎分级的精确性和全面性。方法：提出并设计了牡蛎自动化分级设备,确定了旋转滚筒与挡板传送带结合的牡蛎排队结构、质量检测和机器视觉检测相结合的分级方式,完成了牡蛎分级设备的整体结构设计。通过工业相机采集牡蛎图像,使用大津法二值化、高斯滤波处理、Canny算子边缘提取等方法提取牡蛎图像,通过机器视觉算法以长度和饱满度为标准对牡蛎进行分级,并进行机器视觉分级与人工分级对比试验。结果：该设备分级准确率为95.4％,图像检测速度约为0.647 s/幅。结论：机器视觉对牡蛎分级是有效的,可以较为准确地对牡蛎进行分级。     

Computer vision-based evaluation of pre- and postrigor changes in size and shape of Atlantic cod (Gadus morhua) and Atlantic salmon (Salmo salar) fillets during rigor mortis and ice storage: effects of perimortem handling stress

The present study describes the possibilities for using computer vision-based methods for the detection and monitoring of transient 2D and 3D changes in the geometry of a given product. The rigor contractions of unstressed and stressed fillets of Atlantic salmon (Salmo salar) and Atlantic cod (Gadus morhua) were used as a model system. Gradual changes in fillet shape and size (area, length, width, and roundness) were recorded for 7 and 3 d, respectively. Also, changes in fillet area and height (cross-section profiles) were tracked using a laser beam and a 3D digital camera. Another goal was to compare rigor developments of the 2 species of farmed fish, and whether perimortem stress affected the appearance of the fillets. Some significant changes in fillet size and shape were found (length, width, area, roundness, height) between unstressed and stressed fish during the course of rigor mortis as well as after ice storage (postrigor). However, the observed irreversible stress-related changes were small and would hardly mean anything for postrigor fish processors or consumers. The cod were less stressed (as defined by muscle biochemistry) than the salmon after the 2 species had been subjected to similar stress bouts. Consequently, the difference between the rigor courses of unstressed and stressed fish was more extreme in the case of salmon. However, the maximal whole fish rigor strength was judged to be about the same for both species. Moreover, the reductions in fillet area and length, as well as the increases in width, were basically of similar magnitude for both species. In fact, the increases in fillet roundness and cross-section height were larger for the cod. We conclude that the computer vision method can be used effectively for automated monitoring of changes in 2D and 3D shape and size of fish fillets during rigor mortis and ice storage. In addition, it can be used for grading of fillets according to uniformity in size and shape, as well as measurement of fillet yield measured in thickness. The methods are accurate, rapid, nondestructive, and contact-free and can therefore be regarded as suitable for industrial purposes.     

Investigations into the accuracy of prediction of beef carcass composition using subcutaneous fat thickness and carcass weight II. Improving the accuracy of prediction

Attempts were made to improve the accuracy of prediction of carcass components in 68 steer carcasses using a subcutaneous fat thickness measurement (FTP(8)) and hot side weight as the principal or sole predictors. When carcasses were divided into two weight ranges, prediction of the percentages of carcass fat (fat %) and muscle (muscle %) in the lighter group, using FTP(8), were almost as accurate (fat %: RSD, 1.83; R(2), 0.73; muscle %: RSD, 1.70; R(2), 0.47) as those given by multiple regression measurements made by advanced technologies. Prediction in the heavier weight group was unsatisfactory, featuring high RSDs and a low R(2). Because breed differences in prediction did not occur in the lighter group, a single prediction equation could be used for each carcass component. The use of fat thickness groups gave a similar result to that of weight groups. Once weight groups were used, curvilinear analysis did not improve the accuracy of prediction. In the lighter carcasses, the addition of hot side weight, eye muscle area and muscle score, singly or in combination, did not improve the accuracy of percentage predictions. In the heavier carcasses, the addition of hot side weight and eye muscle area was necessary to give prediction of a modest level of accuracy. For the prediction of the weights of carcass components, the addition of hot side weight to FTP(8) gave the most accurate prediction in light carcasses (fat weight: RSD, 1.60; R(2), 0.92; muscle weight: RSD, 2.08; R(2), 0.97). In the heavy carcasses, hot side weight and eye muscle area added to FTP(8) gave the most accurate prediction (fat weight: RSD, 4.40; R(2), 0.69; muscle weight: RSD, 3.73; R(2), 0.88). Because of the high level of variance explained, it is recommended that weights of carcass components be predicted, particularly in heavy carcasses.     

天然鱼翅与仿生鱼翅快速鉴别方法的建立

为了提高真假鱼翅鉴别的准确性、可操作性及快速性,本文采用反复热冷处理及组织结构对天然鱼翅及仿生鱼翅进行了鉴别方法的研究。反复冷热处理结果发现天然鱼翅均具有明显的反复热收缩和冷伸展的特点,其中最佳条件为热处理5 min、冷处理30 min及反复冷热处理2次。为了进一步验证热冷处理条件的可行性,通过对鲨鱼不同部位及不同种类的鱼翅的热缩冷胀结果进行统计分析,所得的数据密集度好,误差小,准确性强,均值具有代表性,符合正态分布,并且发现仿生鱼翅是热胀冷缩的特点。组织形态结果发现天然鱼翅的翅针表现为不规则的同心环状,环带排列相对均匀,具有鱼类特有的年轮特征,但是仿生鱼翅无年轮特征,两者之间的差异非常明显。总的来说,热缩冷胀法及组织形态法是一种快速的、操作简单的、准确的天然鱼翅与仿生鱼翅的鉴别方法。     

Analysis of image-based measurements and USDA characteristics as predictors of beef lean yield

This paper describes a comprehensive analysis of the 12th rib image measurements, USDA yield characteristics and USDA yield grade as predictors of beef lean yield. The predictors were used in separate groups to construct three sets of multiple linear regression models for the prediction of lean yield of 241 carcasses. Fat thickness is traditionally considered the most useful predictor. The analysis showed that the percent rib eye area was a more useful single predictor than fat thickness, and that the average fat thickness was insignificant when rib eye area and fat area were used. While marbling is traditionally considered a predictor of quality, the results showed that marbling characteristics were also useful for yield prediction. The usually-observed large differences in accuracy between the predictions of lean weight and lean percentage were shown to result from the variations in carcass weight. Statistical diagnostics confirmed the suitability of the models developed. A neural network model was tested and the results suggested that the inclusion of nonlinearity in the predictive models did not prove beneficial. Important predictors were identified and the advantages of computer vision and image processing techniques were further demonstrated.     

Adaptive detection of weft‐knitted fabric defects based on machine vision system

This paper describes a machine vision system for the detection of weft‐knitted fabric defects based on an adaptive pulse‐coupled neural network (PCNN) and Ridgelet transform. In order to classify defects according to their different texture features, two methods are implemented: an improved PCNN method to segment the defects such as hole and dropped stitch from background image and a Ridgelet transform method based on wavelet analysis to identify the defect such as course mark. In implementing the PCNN model, necessary parameters of PCNN model such as linking coefficient, connection weight, and iteration number are automatically calculated in accordance with the spatial distance of neurons, mean, and variance value of whole image, and the cross‐entropy criterion. The function of Ridgelet transform is to identify the straight line marks and fit the regression equation for simulating the course mark in the image. The Ridgelet transform model can be simplified as the combination of Radon and wavelet transforms. The parameters of detected line are acquired by wavelet analysis in Fourier semicircle region. The experiment materials were several plain and interlocked weft‐knitted fabrics with hole, dropped stitch, and course mark defects. The fabric images were captured by an area‐scan camera with a resolution of 600 × 800 pixels, and signal processing was controlled by a digital signal processing multiprocessor on the inspection machine. The validation tests proved that the system performed well.     

The value of in vivo real time ultrasonography in assessing loin muscularity and carcass composition of rabbits

Sixty nine growing rabbits were scanned over the lumbar region using a real time ultrasonography (RTU) machine to estimate loin muscularity and carcass composition. Longissimus thoracis et lumborum muscle (LM) depth, width and area were taken. Animals were weighed (LW), slaughtered and carcass composition was determined. Equivalent measurements to those taken by RTU in vivo were taken on the carcass and muscularity indices were calculated on carcass and in vivo. Simple correlations between the two types of measurements were determined and carcass composition was estimated by simple and multiple regressions. The LW varied from 1200 to 3410g. The simple correlations between carcass and in vivo RTU LM measurements were high (P<0.001) and the LM area was the trait with the highest correlation (r=0.92). Simple correlations between muscularity indices measured by RTU and in carcass were significant (P<0.001). In vivo RTU measurements explained a large amount of the variation of the carcass meat weight (MW) and bone weight (r(2) range from 0.49 to 0.77; P<0.001). Using multiple regression equations to estimate carcass composition, the best fit was obtained with the LW and one or more in vivo RTU measurement. The LW explained 90.6% of the variation of MW in the carcass. In vivo RTU is able to estimate loin muscularity and carcass composition of rabbits with accuracy. The usefulness of in vivo RTU and LW to predict carcass composition of rabbits using multiple regressions was also shown.     

Data Fusion for Food Authentication: Fresh/Frozen–Thawed Discrimination in West African Goatfish (Pseudupeneus prayensis) Fillets

The substitution of fresh fish with frozen–thawed fish is a typical fraud that not only damages consumers from an economical point of view, but also causes safety issues. Furthermore, fish authentication is important for correct product labeling, as promoted by recent regulatory actions. In this paper, we present the results of the authentication of fresh west African goatfish (Pseudupeneus prayensis) fillets using different analytical technologies, namely a portable visible/near-infrared spectrometer, a compact digital camera, and a texture analyzer. First, the classification performance of the abovementioned analytical technologies is evaluated and compared. Then, it is shown how the fusion of different technologies can be effectively used to improve the classification accuracy. Particularly, spectra and color features extracted from digital images returned a classification accuracy of 100 and 98.5 %, respectively, when considered separately. However, the classification accuracy fell to 80 % when considering measurements taken with a 24-h delay. Data fusion, instead, allowed a classification accuracy of 100 % also after 24 h. Hence, the combination of a spectrometer and a digital camera is very promising for cost-effective on-line/at-line applications, as both sensors are rapid, non-invasive, and do not require sample preparation. Furthermore, since more than 200 samples were collected over a prolonged period of time (1 year), the classification models encompassed some sources of variability (seasonality effects and size) that are not usually accounted for in similar studies.