CN112798539A - Intelligent aflatoxin detection method based on transfer learning - Google Patents
Intelligent aflatoxin detection method based on transfer learning Download PDFInfo
- Publication number
- CN112798539A CN112798539A CN202011433944.XA CN202011433944A CN112798539A CN 112798539 A CN112798539 A CN 112798539A CN 202011433944 A CN202011433944 A CN 202011433944A CN 112798539 A CN112798539 A CN 112798539A
- Authority
- CN
- China
- Prior art keywords
- aflatoxin
- pixel
- network
- hyperspectral
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses an aflatoxin intelligent detection method based on transfer learning, which comprises 4 steps: (1) firstly, training a deep learning network by using a public remote sensing hyperspectral data set, (2) finely tuning the network by using a marked aflatoxin standard sample hyperspectral image, (3) identifying the aflatoxin seed hyperspectral image to be detected pixel by the finely tuned network, and (4) visualizing a seed pollution area and inverting the aflatoxin content. The method combines deep migration learning and hyperspectral imaging, migrates the aflatoxin detection data through remote sensing data, detects the aflatoxin by pixel, realizes quantitative detection of the aflatoxin of the seeds, and has great application value.
Description
Technical Field
The invention discloses an aflatoxin intelligent detection method based on transfer learning, and particularly relates to a pixel-level aflatoxin kernel quantitative inversion method combining transfer learning and hyperspectral imaging.
Background
Aflatoxin (Aflatoxin) is a highly toxic and strong carcinogen, has 68 times of toxicity of arsenic, is the strongest (class I) chemical carcinogen found at present, is the largest cause of malignant tumors (particularly liver cancer), and widely exists in peanuts, corns and products thereof. Although the biochemical sampling inspection method represented by liquid chromatogram has high detection precision, the detection speed is slow, and online detection cannot be realized. The aflatoxin has ultraviolet fluorescence characteristics and superficial surface distribution characteristics. The spectrum is a direct reaction of substances, and the online detection of the aflatoxin becomes possible due to the capacity of integrating the hyperspectral images with the map.
However, the traditional method only collects a few key wavelengths and uses the average spectrum of grains for detection due to the limitation of hardware and algorithm, the detection accuracy is low, and the full spectrum detection method for deep learning becomes possible along with the huge breakthrough of GPU hardware and deep learning algorithm. The invention provides a method for quantitatively inverting aflatoxin based on full spectrum, which performs transfer learning through hyperspectral images to obtain a pixel-by-pixel detection model.
Disclosure of Invention
The invention mainly aims to overcome the defects of the problems and discloses an aflatoxin intelligent detection method based on transfer learning.
The above purpose is realized by the following technical scheme:
the method comprises 4 steps: 1) firstly, training a deep learning network by using a public remote sensing hyperspectral data set, 2) finely adjusting the network by using a marked aflatoxin standard sample hyperspectral, 3) then identifying pixels by pixels of a to-be-detected aflatoxin grain hyperspectral image by the finely adjusted network, and 4) visualizing a grain pollution area and inverting the aflatoxin content.
Wherein:
(1) the method comprises the steps of firstly downloading remote sensing hyperspectral images such as Botswana, Pavia and Pavia University hyperspectral datasets from a public dataset, then converting the spectra into images reconstructed by single pixel spectra by pixel using Reshape transformation one by one, inputting the images reconstructed by the single pixel spectra into a mainstream neural network architecture as a training sample set for training, and inputting the images reconstructed by the mainstream neural network architecture such as VGGNet and GoogleNet frameworks, so that the neural network architecture with the spectrum recognition function is automatically built, and the network with the highest accuracy is preferably selected as an original network.
(2) The method comprises the steps of firstly preparing an aflatoxin pollution standard sample, then collecting hyperspectral images of the aflatoxin standard culture, converting the spectra into images reconstructed by single pixel spectra by using Reshape conversion pixel by pixel, inputting the images into an original network as a training sample set, transferring network weights of layers 1 to N-3 of an original neural network to a new network, retraining layers 1 to 3 in the reverse number, changing the number of layers of an activation function and a full connection layer to obtain a target network, and optimizing the target network so that the target network can identify aflatoxin areas with different concentrations.
(3) The finely adjusted network identifies the aflatoxin kernel hyperspectral images pixel by pixel, which means that the aflatoxin-polluted agricultural product kernels are subjected to hyperspectral imaging, and particularly, the special agricultural product kernels such as peanuts, corn kernels and the like are also suitable for Reshape transformation to convert each pixel into a pixel image and input the pixel image into a target network for identification, and whether the pixel is polluted by aflatoxin or not and the pollution degree is judged.
(4) Visualization of a grain pollution area and inversion of aflatoxin content refer to firstly using a heat image to mark aflatoxin pollution areas of different degrees on the grain area image in situ, and then performing inversion with the chemically measured grain aflatoxin content according to the accumulated degree of pollution of each point to obtain the aflatoxin content of grains, wherein a liquid chromatography can be adopted in a special chemical measurement process.
(5) The marking process includes adhering seed and seed coat, especially peanut red coat, to Teflon board with double-sided adhesive, adhering black rubber layer with netted structure to the upper part of the board without fluorescent material, and dropping aflatoxin acetonitrile solution in different grids for use after volatilizing acetonitrile.
The invention has the following effects: the quantitative detection method of the aflatoxin is realized by the patented technology, the online sorting of the aflatoxin seeds can be realized by combining the embedded light path design and the color sorter mechanical device, and the sorting equipment has positive significance for prenatal detoxification of large-scale grain and oil processing enterprises, large-scale feed processing enterprises and large-scale grain and oil foreign trade enterprises, and has wide application prospect. In addition, through the combination of deep learning, transfer learning, hyperspectral imaging and artificial intelligence, the technology content is high, the added value of agricultural products in China is improved, and the social benefit is remarkable.
Drawings
FIG. 1 is a general schematic diagram of an aflatoxin intelligent detection method based on transfer learning.
Detailed Description
Specific embodiments of the present apparatus and method are described below in conjunction with the appended drawings.
In the case of the example 1, the following examples are given,
referring to fig. 1, the general schematic diagram of the intelligent aflatoxin detection method of the present invention is shown, and the method comprises 4 steps:
(1) the method comprises the steps of firstly downloading remote sensing hyperspectral images such as Botswana, Pavia and Pavia University hyperspectral datasets from a public dataset, then converting the spectra into images reconstructed by single pixel spectra by pixel using Reshape transformation one by one, inputting the images reconstructed by the single pixel spectra into a mainstream neural network architecture as a training sample set for training, and inputting the images reconstructed by the mainstream neural network architecture such as VGGNet and GoogleNet frameworks, so that the neural network architecture with the spectrum recognition function is automatically built, and the network with the highest accuracy is preferably selected as an original network.
(2) The method comprises the steps of firstly preparing an aflatoxin pollution standard sample, then collecting hyperspectral images of the aflatoxin standard culture, converting the spectra into images reconstructed by single pixel spectra by using Reshape conversion pixel by pixel, inputting the images into an original network as a training sample set, transferring network weights of layers 1 to N-3 of an original neural network to a new network, retraining layers 1 to 3 in the reverse number, changing the number of layers of an activation function and a full connection layer to obtain a target network, and optimizing the target network so that the target network can identify aflatoxin areas with different concentrations.
(3) The finely adjusted network identifies the aflatoxin kernel hyperspectral images pixel by pixel, which means that the aflatoxin-polluted agricultural product kernels are subjected to hyperspectral imaging, and particularly, the special agricultural product kernels such as peanuts, corn kernels and the like are also suitable for Reshape transformation to convert each pixel into a pixel image and input the pixel image into a target network for identification, and whether the pixel is polluted by aflatoxin or not and the pollution degree is judged.
(4) Visualization of a grain pollution area and inversion of aflatoxin content refer to firstly using a heat image to mark aflatoxin pollution areas of different degrees on the grain area image in situ, and then performing inversion with the chemically measured grain aflatoxin content according to the accumulated degree of pollution of each point to obtain the aflatoxin content of grains, wherein a liquid chromatography can be adopted in a special chemical measurement process.
The aflatoxin standard culture preparation process comprises the following steps: firstly, a seed coat, such as peanut red coat, is stuck on a Teflon plate by using a double-sided adhesive tape, then a grid-shaped structure black rubber layer is stuck on the upper part by using a non-fluorescent material, and finally, aflatoxin acetonitrile solutions with different concentrations are dropped into different grids to be used after the acetonitrile is volatilized.
In order to ensure that two surfaces of the seed are shot completely as much as possible, hyperspectral images can be shot on the positive side and the negative side of the seed, so that the result is more accurate, and the light path system can be embedded into the existing color sorter to replace the original light path, so that the sorter has the aflatoxin detection function.
The invention can detect not only the aflatoxin-polluted corn kernels, but also the other nut-type toxin-polluted kernels, can be applied to quantitative detection of corn and peanut products, and can also be applied to the detection of other objects, as long as the basic idea of the invention is not violated, and the invention is also suitable for the protection scope, and the protection scope is defined by the claims.
Claims (6)
1. An aflatoxin intelligent detection method based on transfer learning is characterized by comprising the following steps: the method comprises 4 steps: (1) firstly, training a deep learning network by using a public remote sensing hyperspectral data set, (2) finely tuning the network by using a marked aflatoxin standard sample hyperspectral image, (3) identifying the aflatoxin seed hyperspectral image to be detected pixel by the finely tuned network, and (4) visualizing a seed pollution area and inverting the aflatoxin content.
2. The aflatoxin intelligent detection method based on transfer learning of claim 1, which is characterized in that: the method comprises the steps of firstly downloading remote sensing hyperspectral images such as Botswana, Pavia and Pavia University hyperspectral data sets from a public data set, then converting the spectrums into images reconstructed by single pixel spectrums by pixel-by-pixel Reshape conversion, inputting the images reconstructed by a mainstream neural network framework such as VGGNet and GoogleNet framework and the like as a training sample set to train, thereby automatically building a neural network framework with a spectrum identification function, preferably selecting a network with the highest accuracy as an original network, and using the special Reshape conversion, namely filling spectrum values on pixel points of a full-0 image line by line, supplementing 0 in insufficient places, wherein the Reshape conversion can be replaced by a plot image or an autocorrelation matrix.
3. The aflatoxin intelligent detection method based on transfer learning of claim 1, which is characterized in that: the method comprises the steps of firstly preparing an aflatoxin pollution standard sample, then collecting hyperspectral images of the aflatoxin standard culture, converting the spectra into images reconstructed by single pixel spectra by pixel-by-pixel Reshape conversion, inputting the images into an original network as a training sample set, transferring network weights of layers 1 to N-3 of the original neural network to a new network, retraining layers 1 to 3 in the reverse number, changing an activation function and the number of layers of a full connection layer to obtain a target network, and optimizing the target network so that the target network can identify aflatoxin areas with different concentrations.
4. The aflatoxin intelligent detection method based on transfer learning of claim 1, which is characterized in that: the finely adjusted network identifies the aflatoxin kernel hyperspectral images pixel by pixel, namely the aflatoxin-polluted agricultural product kernels are subjected to hyperspectral imaging, particularly the agricultural product kernels such as peanuts, corn kernels and the like, and the method is also suitable for Reshape transformation to convert each pixel into a pixel image and input the pixel image into a target network for identification, and whether the pixel is polluted by the aflatoxin or not and the pollution degree of the pixel are judged.
5. The aflatoxin intelligent detection method based on transfer learning of claim 1, which is characterized in that: the visualization of the grain pollution area and the inversion of the aflatoxin content mean that firstly, the heat image is used for marking the aflatoxin pollution areas with different degrees on the grain area image in situ, and then the inversion is carried out according to the accumulation of the pollution degree of each point and the chemically measured grain aflatoxin content to obtain the aflatoxin content of grains, and the liquid chromatography can be adopted in the special chemical measurement process.
6. The aflatoxin intelligent detection method based on transfer learning of claim 3, which is characterized in that: the marking process includes adhering one seed and seed coat, especially peanut red coat, to Teflon board with double-sided adhesive, adhering one black rubber layer of netted structure with non-fluorescent material, dropping aflatoxin acetonitrile solution in different grids for use after volatilizing acetonitrile.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011433944.XA CN112798539A (en) | 2020-12-10 | 2020-12-10 | Intelligent aflatoxin detection method based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011433944.XA CN112798539A (en) | 2020-12-10 | 2020-12-10 | Intelligent aflatoxin detection method based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112798539A true CN112798539A (en) | 2021-05-14 |
Family
ID=75806372
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011433944.XA Pending CN112798539A (en) | 2020-12-10 | 2020-12-10 | Intelligent aflatoxin detection method based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112798539A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113433076A (en) * | 2021-05-18 | 2021-09-24 | 中国检验检疫科学研究院 | Hyperspectral imaging technology-based method for identifying aflatoxin in corn seeds |
| CN114387258A (en) * | 2022-01-14 | 2022-04-22 | 北京理工大学重庆创新中心 | Hyperspectral image reconstruction method based on regional dynamic depth expansion neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120061586A1 (en) * | 2010-09-10 | 2012-03-15 | Haibo Yao | Method and detection system for detection of aflatoxin in corn with fluorescence spectra |
| CN103575694A (en) * | 2013-11-11 | 2014-02-12 | 东北农业大学 | Quick detection method for aflatoxin B1 of rice based on near infrared spectroscopy |
| CN104646315A (en) * | 2015-03-02 | 2015-05-27 | 青岛农业大学 | Intelligent agricultural product sorting machine with aflatoxin detection function |
| CN107576618A (en) * | 2017-07-20 | 2018-01-12 | 华南理工大学 | Pyricularia Oryzae detection method and system based on depth convolutional neural networks |
| CN111914696A (en) * | 2020-07-16 | 2020-11-10 | 河海大学 | Hyperspectral remote sensing image classification method based on transfer learning |
-
2020
- 2020-12-10 CN CN202011433944.XA patent/CN112798539A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120061586A1 (en) * | 2010-09-10 | 2012-03-15 | Haibo Yao | Method and detection system for detection of aflatoxin in corn with fluorescence spectra |
| CN103575694A (en) * | 2013-11-11 | 2014-02-12 | 东北农业大学 | Quick detection method for aflatoxin B1 of rice based on near infrared spectroscopy |
| CN104646315A (en) * | 2015-03-02 | 2015-05-27 | 青岛农业大学 | Intelligent agricultural product sorting machine with aflatoxin detection function |
| CN107576618A (en) * | 2017-07-20 | 2018-01-12 | 华南理工大学 | Pyricularia Oryzae detection method and system based on depth convolutional neural networks |
| CN111914696A (en) * | 2020-07-16 | 2020-11-10 | 河海大学 | Hyperspectral remote sensing image classification method based on transfer learning |
Non-Patent Citations (3)
| Title |
|---|
| HAN ZHONGZHI等: "Pixel-level aflatoxin detecting based on deep learning and hyperspectral imaging", 《COMPUTERS AND ELECTRONICS IN AGRICULTURE》 * |
| ZHU HONGFEI等: "Identifying carrot appearance quality by transfer learning", 《JOURNAL OF FOOD PROCESS ENGINEERING》 * |
| 韩仲志等: "高光谱亚像元分解预测花生中的黄曲霉毒素B1", 《中国食品学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113433076A (en) * | 2021-05-18 | 2021-09-24 | 中国检验检疫科学研究院 | Hyperspectral imaging technology-based method for identifying aflatoxin in corn seeds |
| CN114387258A (en) * | 2022-01-14 | 2022-04-22 | 北京理工大学重庆创新中心 | Hyperspectral image reconstruction method based on regional dynamic depth expansion neural network |
| CN114387258B (en) * | 2022-01-14 | 2024-03-22 | 北京理工大学重庆创新中心 | Hyperspectral image reconstruction method based on regional dynamic depth expansion neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Barman et al. | Soil texture classification using multi class support vector machine | |
| Wang et al. | A robust Multi-Band Water Index (MBWI) for automated extraction of surface water from Landsat 8 OLI imagery | |
| Zhi et al. | Using transfer learning with convolutional neural networks to diagnose breast cancer from histopathological images | |
| Luo et al. | PM2. 5 concentration estimation using convolutional neural network and gradient boosting machine | |
| CN107421894A (en) | Based on unmanned plane EO-1 hyperion inverting heavy metal in soil pollution monitoring method | |
| Riid et al. | Pavement distress detection with deep learning using the orthoframes acquired by a mobile mapping system | |
| CN112798539A (en) | Intelligent aflatoxin detection method based on transfer learning | |
| CN107665492A (en) | Colon and rectum panorama numeral pathological image tissue segmentation methods based on depth network | |
| Yu et al. | An integrated rice panicle phenotyping method based on X-ray and RGB scanning and deep learning | |
| CN101266258A (en) | Turbulence condition low concentration sludge settlement rate measuring method and the measuring device | |
| Lyu et al. | Extracting the tailings ponds from high spatial resolution remote sensing images by integrating a deep learning-based model | |
| CN101477025B (en) | Fast evaluation method for collection exhibition materials based on image processing | |
| CN115187865B (en) | Unmanned aerial vehicle image near-infrared band simulation method and medium based on ground object spectrum | |
| Mengistu et al. | Soil Characterization and Classification: A Hybrid Approach of Computer Vision and Sensor Network. | |
| CN111222545B (en) | Image classification method based on linear programming incremental learning | |
| CN112489026A (en) | Asphalt pavement disease detection method based on multi-branch parallel convolution neural network | |
| Ainiwaer et al. | Deep learning-based rapid recognition of oasis-desert ecotone plant communities using UAV low-altitude remote-sensing data | |
| Han et al. | Vision-based crack detection of asphalt pavement using deep convolutional neural network | |
| Huang et al. | A fast antibiotic detection method for simplified pretreatment through spectra-based machine learning | |
| Song et al. | Assessment of wheat chlorophyll content by the multiple linear regression of leaf image features | |
| Murphy et al. | Spatial variation of chlorophyll on estuarine mudflats determined by field-based remote sensing | |
| CN107576600A (en) | A kind of quick determination method for smearing tea grain size category | |
| Shafi et al. | Embedded AI for wheat yellow rust infection type classification | |
| CN107044947B (en) | A kind of recognition methods of the PM2.5 pollution index based on characteristics of image | |
| Pandiri et al. | Smart soil image classification system using lightweight convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210514 |