CN101943661B - Near infrared spectrum and microscopic bacterial plaque area data fusion-based pork freshness non-destructive testing technology - Google Patents

Abstract

Translated fromChinese

一种肉类质量检测系统，包括采集肉类样品脂肪组织图像的图像采集模块、提取所述脂肪组织图像中的培养基的图像处理模块、获取所述培养基中的细菌菌斑的总面积与培养基面积的比值的特征值提取模块、解析所述样品的近红外光谱以获取所述肉类样品的结构组成数据的光谱分析模块、及对所述特征值提取模块获取的比值和所述肉类样品的结构组成数据进行融合处理以得到一实际输出值的数据融合处理模块，所述培养基是在所述脂肪组织图像中检测到的仅具有所述肉类样品的脂肪组织的一固定区域，所述实际输出值表征所述肉类样品的质量。本发明还提供一种肉类质量检测方法，所述肉类质量检测系统及方法可快速、准确检测肉类质量，且结构和检测过程简单。

A meat quality detection system, comprising an image acquisition module for collecting adipose tissue images of meat samples, an image processing module for extracting culture medium in the fat tissue images, obtaining the total area and the total area of bacterial plaque in the culture medium A characteristic value extraction module for the ratio of the culture medium area, a spectral analysis module for analyzing the near-infrared spectrum of the sample to obtain the structural composition data of the meat sample, and the ratio obtained by the characteristic value extraction module and the meat A data fusion processing module that performs fusion processing on the structural composition data of the meat-like sample to obtain an actual output value, the culture medium is a fixed region detected in the fat tissue image that only has the fat tissue of the meat sample , the actual output value characterizes the quality of the meat sample. The invention also provides a meat quality detection method. The meat quality detection system and method can quickly and accurately detect meat quality, and the structure and detection process are simple.

Description

A kind of pork freshness non-destructive testing technology based near infrared spectrum and microscopic bacterial plaque area data fusion

Technical field

The present invention relates to a kind of detection system, relate in particular to a kind of meat quality detection system and method.

Background technology

In recent years, Public Health Emergencies serious threat national health, therefore, sanitary inspection, has especially caused great attention to the quality testing of the meats such as pork, mutton, beef.At present, the detection technique of meat quality has organoleptic detection, microorganism detection and physics and chemistry detection etc.Organoleptic detection dependence testing staff's sense organ judges the freshness of meat, therefore, higher to testing staff's Capability Requirement, must be that people through system training and long-term practice could be competent at, and it is more affected by subjective factor, testing staff's individual sense organ difference directly affects the accuracy of testing result.Microorganism detection and physics and chemistry detection need to rely on a series of chemical devices and complete, and process is complicated, detection time is long, apparatus expensive, can not carry out fast scene and detect in real time.

Summary of the invention

In view of the above-mentioned problems in the prior art, fundamental purpose of the present invention is to address the deficiencies of the prior art, and provides a kind of and can fast, accurately detect meat quality and structure, the simple meat quality detection system of process and method.

A kind of meat quality detection system, the image capture module that comprises a collection meat sample adipose tissue image, one extracts the image processing module of the nutrient culture media in described adipose tissue image, one obtains the eigenwert extraction module of the total area of the bacterium bacterial plaque in described nutrient culture media and the ratio of nutrient culture media area, one near infrared spectrum of resolving described sample forms the spectral analysis module of data to obtain the structure of described meat sample, and the structure of the ratio that obtains of a pair of described eigenwert extraction module and described meat sample forms data and carries out fusion treatment to obtain the Data Fusion module of a real output value, described nutrient culture media is the fixed area detecting in described adipose tissue image, this fixed area only comprises the adipose tissue image of described meat sample, described real output value characterizes the quality of described meat sample.

A meat quality detection method, comprises the following steps:

Acquisition step: the adipose tissue image that gathers a meat sample;

Nutrient culture media extraction step: extract a fixed area of the adipose tissue only with described meat sample in described adipose tissue image as nutrient culture media;

Eigenwert extraction step: calculate the area of the bacterial plaque in described nutrient culture media and the ratio of described nutrient culture media area;

Spectrometry procedure: the structure of obtaining described meat sample by resolving the near infrared spectrum of described meat sample forms data; And

Data fusion step: the structure of the ratio of described bacterial plaque area and nutrient culture media area and described meat sample is formed in the default algorithm of data substitution one and merges computing, to export a real output value, described real output value is in order to characterize the quality of described meat sample.

Described meat quality detection system and method are by gathering the adipose tissue image of described sample and image being processed and extract nutrient culture media from described image, by obtaining the total area of the bacterium bacterial plaque in described nutrient culture media and the ratio of nutrient culture media area as the First Eigenvalue of analyzing the quality of meat to be detected, avoided only calculating the data deviation that amount of bacterial plaque is brought, and using the structure of obtaining described meat sample and form data as the Second Eigenvalue of analyzing meat quality to be detected by resolving the near infrared spectrum of described sample, finally to described first, Second Eigenvalue carries out Data Fusion and exports the data that characterize meat quality to be detected, can be quick, accurately detect meat quality, and detection architecture and process are simple.

Accompanying drawing explanation

Fig. 1 is the block scheme of meat quality detection system better embodiment of the present invention.

Fig. 2 A-2D is respectively sample at the nutrient culture media image of corrupt 0 hour, 6 hours, 12 hours and 24 hours.

Fig. 3 is the structural representation of the neural network unit in Fig. 1.

Fig. 4 is the process flow diagram of meat quality detection method better embodiment of the present invention.

Fig. 5 is that the neural network unit in Fig. 1 obtains the process flow diagram of the data that characterize meat quality to be detected according to first, second eigenwert of described sample.

Embodiment

The invention will be further described below in conjunction with the accompanying drawings and the specific embodiments.

Please refer to Fig. 1 to Fig. 3, the better embodiment of meat quality detection system 1 of the present invention comprises animage capture module 10, animage processing module 20, aneigenwert extraction module 30, aspectral analysis module 40 and a Data Fusion module 50.Describedimage processing module 20 comprises animage conversion unit 22, a nutrient culturemedia extraction unit 24 and animage enhancing unit 26.

Describedimage capture module 10 comprises a CCD (Charge-coupled Device, charge coupled cell) microscope, for gathering the image of meat sample adipose tissue to be detected, in described adipose tissue, contain the quality index that characterizes described meat to be detected, be grown in the bacterial plaque of the bacterium of described meat to be detected.During concrete operations, the scalpel of available sterilization extracts in the adipose tissue of meat to be detected samples long 2 centimetres, wide 1 centimetre, thick 2 millimeters, the sample of extraction is placed on microslide, again microslide is placed on the microscopical objective table of described CCD, adjust the position of objective table, and described CCD microscope is arranged to a suitable amplification coefficient, as 100 times, by described CCD microscope, sample is taken, to obtain RGB (red channel, green channel, the blue channel) image of described adipose tissue.

Because the green channel of image can well react the bacterium bacterial plaque in adipose tissue, the green channel that describedimage conversion unit 22 is extracted in described RGB image, and the image that only has green channel extracting is converted to the gray level image of 8, object in described gray level image also comprises double dish, i.e. described microslide.

Described nutrient culturemedia extraction unit 24 extracts nutrient culture media by Hough transform method from described gray level image, and described nutrient culture media is the border circular areas in described gray level image, and described nutrient culture media is to only include adipose tissue, and does not comprise the image-region of described double dish.According to double dish, nutrient culture media shared position in whole image, if the width that to set the radius of a circle detect by Hough transformation be whole gray level image 1/3rd to 1/2nd between, and the center of circle is when take in the circle of sixth of the width that described gray level image center is whole gray level image as center of circle radius, detected circle is effectively round, extracts the region of the circle of radius minimum in effective circle as described nutrient culture media.In other embodiments, the elliptical region that also can detect in described gray level image by Hough transformation is used as described nutrient culture media, and by Hough transformation, in image, detecting circle and ellipse is the known technology that image is processed, and does not repeat them here.

Describedimage enhancing unit 26 utilizes histogram equalization converter technique the image at described nutrient culture media place to be strengthened to the processing of picture contrast, and be sent to describedeigenwert extraction module 30 by having strengthened the contrast image with nutrient culture media afterwards, the described contrast image afterwards that strengthened is as shown in Fig. 2 A-2D, border circular areas in Fig. 2 A-2D is nutrient culture media, and the spot being dispersed in nutrient culture media is bacterium bacterial plaque.

Describedeigenwert extraction module 30 has strengthened contrast image afterwards by detection and has obtained a First Eigenvalue that characterizes sample freshness, because bacteria flora is that form with bacterial plaque is dispersed on nutrient culture media, and bacterial plaque is not of uniform size, therefore, describedeigenwert extraction module 30 obtains described the First Eigenvalue by calculating the area of the bacterial plaque in described nutrient culture media, and described the First Eigenvalue is the ratio of the total area of all bacterial plaques and the area of whole nutrient culture media in described nutrient culture media.In present embodiment, describedeigenwert extraction module 30 is the quantity of the pixel of statistics nutrient culture media bacterial plaque total areas of obtaining all bacterial plaques, describedeigenwert extraction module 30 comprises apixel detection unit 32, onefirst totalizer 34, onesecond totalizer 36 and anarithmetic element 38, describedpixel detection unit 32 detect strengthened each pixel of image after contrast gray-scale value and the position of the pixel that detects in this image, as shown in Fig. 2 A-2D, black splotch in image represents bacterium bacterial plaque, because the gray-scale value of black picture element in image is 0, when the gray-scale value that a pixel detected when describedpixel detection unit 32 is 0, described thefirst totalizer 34 adds 1, when a pixel being detected and be arranged in described nutrient culture media, described thesecond totalizer 36 adds 1, describedarithmetic element 38 calculates described first, the ratio of the accumulated value of the second totalizer, be the ratio of the bacterial plaque total area and described nutrient culture media area, this ratio is as described the First Eigenvalue, ratio is larger, the degree of spoilage of interpret sample is darker, as shown in Fig. 2 A-2D, be respectively sample at corrupt 0 hour, 6 hours, the nutrient culture media of 12 hours and 24 hours, wherein the ratio of bacterial plaque area and nutrient culture media area increases along with the increase of degree of spoilage.

Describedspectral analysis module 40 is for detection of a Second Eigenvalue of sample, and described Second Eigenvalue comprises that the structure of described sample forms data.Describedspectral analysis module 40 comprises a near-infrared light source 42 and a detectingunit 44, described near-infrared light source 42 irradiates described sample, described detectingunit 44 gathers the near infrared spectrum of described sample, the structure of analyzing meat by resolving the near infrared spectrum of described sample forms, as the basis of judgement Meat.As shown in following table (meat hydric group sum of fundamental frequencies and multiple frequency absorption band composition at different levels form table), because near infrared spectrum district is consistent with the sum of fundamental frequencies of hydric group vibration in organic molecule and the uptake zone of frequencys multiplication at different levels, between the structure composition of sample and near infrared spectrum, there is certain funtcional relationship, by the near infrared spectrum of scanning samples, and according to the funtcional relationship between the structure composition of described sample and near infrared spectrum, utilize stoechiometric process to obtain the characteristic information of organic molecule hydric group in sample, because the composition of meat is (as protein, water, multi-layer biological tissue, protein decomposition etc.) mostly by these hydric groups, formed, the structure that therefore can obtain meat to be detected by resolving the near infrared spectrum of described sample forms data.

Further, in order to reduce the impact of the luminous intensity of near-infrared light source 42 on testing result, described near-infrared light source 42 is by sample described in the near infrared light of two bundle different wave lengths the spectrum of collected specimens under this two different wave length respectively, by calculation sample, to the absorbance of the near infrared light of this two different wave length is poor, is analyzed meat structure to be detected and forms data

Further, described Second Eigenvalue also comprises the content data of each constituent of described sample, describedspectral analysis module 40 also can comprise amid-infrared light source 46, auxiliary described near-infrared light source 42 and described detectingunit 44 carry out the quality testing of meat to be detected, describedmid-infrared light source 46 irradiates described sample, the middle infrared spectrum of described detectingunit 44 collected specimens, utilize the relation of sample to the absorption intensity of the absorption band of mid-infrared light and molecular composition or hydric group content, treat sample and carry out quantitative test, further detect the purity of meat.

DescribedData Fusion module 50 comprises a BP (Back Propagation, backpropagation)neural network unit 52, in describedneural network unit 52, prestore a plurality of training samples and corresponding desired output, each training sample comprises that a ratio data and corresponding sample structure form and component content data, each desired output is the value calculating in the default algorithm of the data substitution in corresponding training sample one, represent the quality of different samples, as shown in Figure 3, describedneural network unit 52 has the input layer A of a reception input parameter, one carries out the hidden layer B of data processing and the output layer C of an Output rusults, when detecting the quality of described meat to be detected, the input layer A of institute'sneural network unit 52 receives first of described sample, Second Eigenvalue, and at described hidden layer B by described first, described in Second Eigenvalue substitution, in default algorithm, merge computing, to obtain a real output value, find with receive first, the immediate training sample of Second Eigenvalue, union is to desired output that should training sample and the error amount between described real output value, whether the error amount described in identification between real output value and described desired output is in the error range of a permission, if the error amount between described real output value and described desired output is in the error range of described permission, , by described output layer C, export described real output value, in order to characterize the quality of described meat to be detected, if the error between described real output value and described desired output is outside the error range of described permission, describedneural network unit 52 changes the weights of described default algorithm at its hidden layer B, again according to described first, second eigenwert, calculate real output value, until the error amount between the real output value calculating and described desired output is in the error range of described permission, then export described real output value by described output layer C.

Please continue to refer to Fig. 4, when the better embodiment of meat quality detection method of the present invention utilizes described meat quality detection system 1 to detect the quality of described meat to be detected, comprise the following steps:

Step S1: describedimage capture module 10 is by sample described in described CCD microscope photographing, to obtain the RGB image of adipose tissue in described sample;

Step S2: the green channel that describedimage conversion unit 22 is extracted in described RGB image, and the image that only has green channel extracting is converted to the gray level image of 8;

Step S3: described nutrient culturemedia extraction unit 24 extracts nutrient culture media by Hough transform method from described gray level image, described nutrient culture media is a border circular areas or the elliptical region in described gray level image, described nutrient culture media is to only include adipose tissue image, and does not comprise the circular or oval-shaped fixed area of described double dish image.

Step S4: describedimage enhancing unit 26 utilizes histogram equalization converter technique to strengthen the contrast of described nutrient culture media place image;

Step S5: the area of bacterial plaque and the ratio of described nutrient culture media area that describedeigenwert extraction module 30 calculates in described nutrient culture media, as described the First Eigenvalue, describedeigenwert extraction module 30 is the quantity of the pixel of statistics nutrient culture media bacterial plaque total areas of obtaining all bacterial plaques;

Step S6: described near-infraredspectrum analysis module 40 is obtained sample structure by resolving the near infrared spectrum of described sample forms data, as described Second Eigenvalue.This step is to the absorbance of the near infrared light of two different wave lengths is poor, to come analytic sample structure to form by calculation sample.DescribedInfrared spectroscopy module 40 is also by the middle infrared spectrum of collected specimens, utilize sample the relation of the absorption intensity of the absorption band of mid-infrared light and molecular composition or hydric group content to be detected to the content data of each constituent of sample, now, described Second Eigenvalue also comprises the content data of each constituent of described sample; And

Step S7: describedneural network unit 52 is by merging computing in algorithm default described in described first, second eigenwert substitution, to export a real output value, in order to characterize the quality of described meat to be detected.

Please continue to refer to Fig. 5, described step S7 comprises the following steps:

Step S71: the input layer A of describedneural network unit 52 receives described first, second eigenwert;

Step S72: at described hidden layer B by merging computing in algorithm default described in described first, second eigenwert substitution, to obtain a real output value;

Step S73: find and the immediate training sample of first, second eigenwert receiving, union is to desired output that should training sample and the error amount between described real output value;

Step S74: judge that error amount between described real output value and described desired output is whether in the error range of a permission, if the error amount between described real output value and described desired output is in the error range of described permission, perform step S75, otherwise, execution step S76;

Step S75: export described real output value by described output layer C; And

Step S76: the weights that change described default algorithm, and return to step S71 and again according to described first, second eigenwert, calculate real output value, until the error amount between the real output value calculating and described desired output is in the error range of described permission, then carry out described step S75.

Described meat quality detection system and method are by gathering the adipose tissue image of described sample and image being processed and extract nutrient culture media from described image, by obtaining the total area of the bacterium bacterial plaque in described nutrient culture media and the ratio of nutrient culture media area as the First Eigenvalue of analyzing the quality of meat to be detected, avoided only calculating the data deviation that amount of bacterial plaque is brought, and using the structure of obtaining described meat sample and form data as the Second Eigenvalue of analyzing meat quality to be detected by resolving the near infrared spectrum of described sample, finally to described first, Second Eigenvalue carries out Data Fusion and exports the data that characterize meat quality to be detected, can be quick, accurately detect meat quality, and detection architecture and process are simple.

Claims (8)

Translated fromChinese

1.一种肉类质量检测系统，包括一采集肉类样品脂肪组织图像的图像采集模块、一提取所述脂肪组织图像中的培养基的图像处理模块、一获取所述培养基中的细菌菌斑的总面积与培养基面积的比值的特征值提取模块、一解析所述样品的近红外光谱以获取所述肉类样品的结构组成数据的光谱分析模块、及一对所述特征值提取模块获取的比值和所述肉类样品的结构组成数据进行融合处理以得到一实际输出值的数据融合处理模块，所述光谱分析模块包括近红外光源和检测单元，所述近红外光源通过两束不同波长的近红外光照射所述肉类样品，所述检测单元采集所述肉类样品在所述两不同波长下的光谱，并通过计算所述肉类样品对所述两不同波长的近红外光的吸收度差来分析所述肉类样品的结构组成数据；所述培养基是在所述脂肪组织图像中检测到的一固定区域，该固定区域仅包括所述肉类样品的脂肪组织图像，所述实际输出值表征所述肉类样品的质量。1. A meat quality detection system, comprising an image acquisition module for collecting adipose tissue images of meat samples, an image processing module for extracting the culture medium in the fat tissue images, and an image processing module for obtaining bacterial bacteria in the culture medium A characteristic value extraction module for the ratio of the total area of the spot to the culture medium area, a spectral analysis module for analyzing the near-infrared spectrum of the sample to obtain the structural composition data of the meat sample, and a pair of the characteristic value extraction modules The obtained ratio and the structural composition data of the meat sample are fused to obtain a data fusion processing module for an actual output value. The spectral analysis module includes a near-infrared light source and a detection unit, and the near-infrared light source passes through two different The near-infrared light of the wavelength irradiates the meat sample, and the detection unit collects the spectra of the meat sample at the two different wavelengths, and calculates the response of the meat sample to the near-infrared light of the two different wavelengths. to analyze the structural composition data of the meat sample; the medium is a fixed region detected in the adipose tissue image, and the fixed region only includes the adipose tissue image of the meat sample, The actual output value characterizes the quality of the meat sample.2.如权利要求1所述的肉类质量检测系统，其特征在于，所述图像处理模块包括一提取所述肉类样品脂肪组织图像中的绿色通道并将提取的图像转换为8位灰度图像的图像转换单元、一通过霍夫变换法从所述灰度图像中提取培养基的培养基提取单元、及一增强所述培养基所在图像对比度的图像增强单元。2. meat quality detection system as claimed in claim 1, is characterized in that, described image processing module comprises a green channel that extracts described meat sample adipose tissue image and converts the image that extracts into 8-bit grayscale An image conversion unit for images, a culture medium extraction unit for extracting culture medium from the grayscale image by Hough transform method, and an image enhancement unit for enhancing the contrast of the image where the culture medium is located.3.如权利要求2所述的肉类质量检测系统，其特征在于，所述培养基是在所述灰度图像中提取的一圆形区域，所述圆形区域的半径为整个灰度图像的宽度的三分之一到二分之一之间，所述圆形区域的圆心处于所述灰度图像的一圆内，该圆的圆心为所述灰度图像中心，半径为整个灰度图像的宽度的六分之一。3. The meat quality detection system according to claim 2, wherein the culture medium is a circular area extracted in the grayscale image, and the radius of the circular area is the whole grayscale image Between one-third and one-half of the width, the center of the circular area is within a circle of the grayscale image, the center of the circle is the center of the grayscale image, and the radius is the entire grayscale One-sixth the width of the image.4.如权利要求1所述的肉类质量检测系统，其特征在于，所述特征值提取模块包括一像素检测单元、一第一累加器、一第二累加器及一运算单元，所述像素检测单元检测所述培养基所在图像各像素的灰度值以及所检测像素的位置，当检测到所述培养基所在图像中的一像素的灰度值为0时，所述第一累加器加1，当该像素位于所述培养基中时，所述第二累加器加1，所述运算单元计算所述第一、第二累加器的累加值的比值，即所述菌斑总面积与所述培养基面积的比值。4. meat quality detection system as claimed in claim 1, is characterized in that, described eigenvalue extracting module comprises a pixel detection unit, a first accumulator, a second accumulator and an operation unit, and the pixel The detection unit detects the gray value of each pixel of the image where the culture medium is located and the position of the detected pixel, and when it detects that the gray value of a pixel in the image where the culture medium is located is 0, the first accumulator adds 1. When the pixel is in the culture medium, add 1 to the second accumulator, and the calculation unit calculates the ratio of the accumulated values of the first and second accumulators, that is, the total area of plaque and The ratio of the medium area.5.如权利要求1所述的肉类质量检测系统，其特征在于，所述光谱分析模块还包括一照射所述肉类样品的中红外光源，所述检测单元通过解析所述肉类样品的中红外光谱得到所述肉类样品的各组成成分的含量数据，所述数据融合处理模块对所述特征值提取模块获取的比值、所述光谱分析模块获取的肉类样品的结构组成数据和各组成成分的含量数据进行融合处理以得到所述实际输出值。5. The meat quality detection system according to claim 1, wherein the spectral analysis module also includes a mid-infrared light source that irradiates the meat sample, and the detection unit analyzes the meat sample The content data of each component of the meat sample is obtained by mid-infrared spectroscopy, and the data fusion processing module obtains the ratio obtained by the feature value extraction module, the structural composition data of the meat sample obtained by the spectral analysis module and each The content data of the components are fused to obtain the actual output value.6.一种肉类质量检测方法，包括以下步骤：6. A meat quality detection method, comprising the following steps:采集步骤：采集一肉类样品的脂肪组织图像；Acquisition step: collecting an adipose tissue image of a meat sample;培养基提取步骤：提取所述脂肪组织图像中的仅具有所述肉类样品的脂肪组织的一固定区域作为培养基；medium extraction step: extracting a fixed region of the adipose tissue in the adipose tissue image that only has the meat sample as a medium;特征值提取步骤：计算所述培养基中的菌斑的面积与所述培养基面积的比值；Feature value extraction step: calculating the ratio of the area of plaque in the culture medium to the area of the culture medium;光谱分析步骤：通过解析所述肉类样品的近红外光谱来获取所述肉类样品的结构组成数据，所述光谱分析步骤通过计算所述肉类样品对两种不同波长的近红外光的吸光度差来分析所述肉类样品的结构组成数据；以及Spectral analysis step: Obtain the structural composition data of the meat sample by analyzing the near-infrared spectrum of the meat sample, and the spectral analysis step calculates the absorbance of the meat sample to two different wavelengths of near-infrared light analyzing the structural composition data of the meat sample; and数据融合步骤：将所述菌斑面积与培养基面积的比值以及所述肉类样品的结构组成数据代入一预设的运算法则中进行融合运算，以输出一实际输出值，所述实际输出值用以表征所述肉类样品的质量。Data fusion step: Substituting the ratio of the plaque area to the culture medium area and the structural composition data of the meat sample into a preset algorithm for fusion operation to output an actual output value, the actual output value To characterize the quality of the meat sample.7.如权利要求6所述的肉类质量检测方法，其特征在于，所述采集步骤和培养基提取步骤之间还包括以下步骤：7. meat quality detection method as claimed in claim 6, is characterized in that, also comprises the following steps between described collection step and medium extraction step:提取所述脂肪组织图像中的绿色通道，并将提取的只有绿色通道的图像转换为8位的灰度图像；以及Extracting the green channel in the adipose tissue image, and converting the extracted image with only the green channel into an 8-bit grayscale image; and通过霍夫变换法从所述灰度图像中提取一只包含脂肪组织图像的圆形区域或一椭圆形区域作为所述培养基。A circular region or an elliptical region containing adipose tissue image is extracted from the grayscale image by Hough transform method as the culture medium.8.如权利要求6所述的肉类质量检测方法，其特征在于，所述数据融合步骤包括：8. meat quality detection method as claimed in claim 6, is characterized in that, described data fusion step comprises:通过一神经网络单元的输入层接收所述培养基中的菌斑面积与培养基面积的比值以及所述肉类样品的结构组成数据；receiving the ratio of the plaque area in the culture medium to the culture medium area and the structural composition data of the meat sample through an input layer of a neural network unit;将所述培养基中的菌斑面积与培养基面积的比值以及所述肉类样品的结构组成数据代入一预设的运算法则中进行融合运算，以得到一实际输出值；Substituting the ratio of the plaque area in the culture medium to the culture medium area and the structural composition data of the meat sample into a preset algorithm for fusion calculation to obtain an actual output value;找到与接收的数据最接近的训练样本，并运算对应该训练样本的期望输出值和所述实际输出值之间的误差值，所述训练样本包括一菌斑面积与培养基面积的比值数据和对应的样本结构组成数据，各期望输出值是将对应的训练样本中的数据代入所述预设运算法则中计算出的值，代表不同样本的质量；Find the training sample closest to the received data, and calculate the error value between the expected output value corresponding to the training sample and the actual output value, the training sample includes the ratio data of a plaque area and the culture medium area and The corresponding sample structure constitutes data, and each expected output value is a value calculated by substituting the data in the corresponding training sample into the preset algorithm, representing the quality of different samples;判断所述实际输出值与所述期望输出值之间的误差值是否在一允许的误差范围内；judging whether the error value between the actual output value and the expected output value is within an allowable error range;如果所述实际输出值与所述期望输出值之间的误差值在所述允许的误差范围内，由所述神经网络单元的一输出层输出所述实际输出值；以及if the error value between the actual output value and the expected output value is within the allowable error range, outputting the actual output value by an output layer of the neural network unit; and如果所述实际输出值与所述期望输出值之间的误差值不在所述允许的误差范围内，改变所述预设运算法则的权值，并重新将所述培养基中的菌斑面积与培养基面积的比值以及所述肉类样品的结构组成数据代入所述预设的运算法则中进行融合运算，直至计算得到的实际输出值与所述期望输出值之间的误差值在所述允许的误差范围内，再由所述输出层输出符合条件的实际输出值。If the error value between the actual output value and the expected output value is not within the allowable error range, change the weight of the preset algorithm, and re-calculate the plaque area in the culture medium and The ratio of the medium area and the structural composition data of the meat sample are substituted into the preset algorithm for fusion operation until the error value between the calculated actual output value and the expected output value is within the allowable Within the error range, the output layer outputs the actual output value that meets the conditions.

Images