CN115078327A - Rapid detection method for dangerous chemicals based on neural network architecture search - Google Patents
Rapid detection method for dangerous chemicals based on neural network architecture search Download PDFInfo
- Publication number
- CN115078327A CN115078327A CN202210665879.6A CN202210665879A CN115078327A CN 115078327 A CN115078327 A CN 115078327A CN 202210665879 A CN202210665879 A CN 202210665879A CN 115078327 A CN115078327 A CN 115078327A
- Authority
- CN
- China
- Prior art keywords
- network architecture
- spectral data
- neural network
- data
- dangerous chemicals
- 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
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/65—Raman scattering
-
- 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2201/00—Features of devices classified in G01N21/00
- G01N2201/12—Circuits of general importance; Signal processing
- G01N2201/129—Using chemometrical methods
- G01N2201/1296—Using chemometrical methods using neural networks
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The invention discloses a rapid detection method for dangerous chemicals based on neural network architecture search, which comprises the following steps: acquiring spectral data of dangerous chemicals; step two, designing a network based on the neural network architecture searching method of AutoKeras; step three, obtaining an optimal network architecture; in the first step, for solid chemicals, laser is directly irradiated on the surface of a sample at a focal length of 7.5ms to acquire spectral data, and for liquid chemicals, the sample is required to be arranged in a standard sample bottle to acquire the spectral data; the spectrum of dangerous chemicals is measured by the Raman spectrometer, a neural network architecture searching method is introduced, the AutoKeras framework is adopted for development and design of a self-running network architecture, and compared with other neural network methods, the optimal network architecture is found by searching and integrating network modules, evaluating performance and feeding back, so that the use threshold of the neural network is lowered, and the consumption of time and labor force is reduced.
Description
Technical Field
The invention relates to the technical field of neural networks, in particular to a method for rapidly detecting dangerous chemicals based on neural network architecture search.
Background
The detection methods corresponding to harmful substances such as ammonium nitrate substances, nitroglycerin explosives and the like include X-ray, Gas Chromatography (GC), neutron detection technology and the like, and the detection methods have the problems of complex operation, long detection time and the like; the Raman Spectrum (RS) technology is simple to operate, can distinguish the material type according to the vibration spectrum characteristic of high specificity, has been used for the detection analysis of various harmful substances, however, after the Raman spectrum is obtained by using the Raman spectrum technology to detect dangerous chemicals, the spectrum needs to be preprocessed, cosmic rays, fluorescence effect and other noises are removed by using an upper limit spectrum method, a spectrum filtering method, polynomial transformation, wavelet transformation and the like, but the methods can deform the spectrum and introduce other noises, meanwhile, the characteristic peak matching of the molecular structure needs to be carried out after the operations are completed, and for multi-class spectrum data, a large amount of labor cost and time cost are consumed by a matching method which is compared one by one, matched one by one and determined one by one;
in the prior art, the type and the content of an object to be detected can be automatically and intelligently analyzed by utilizing a traditional machine learning method, although the speed of analyzing spectral data is accelerated to a certain extent, the data processing is too single, the data processing is only effective on specific data, the expansibility is insufficient, the effect is not ideal when various different material types are processed, and the requirements on multi-type analysis cannot be met; the automatic learning feature and the hierarchical feature extraction of deep learning are excellent in classification and regression, and the extensible network structure makes up the defects of machine learning in different data sets, multiple classes and big data application; from VGG-16 to VGG-19, ResNet-50 and ResNet-101, based on simple networks, the network layer is continuously expanded from the aspects of depth, width and resolution of the network, so that more abstract and accurate characteristics can be obtained, and complex problems can be easily processed; however, these significant network models are manually well-designed, such large networks for complex tasks exhaust a lot of material and time costs, and the optimization of network architecture and hyper-parameters is not user-friendly or even professional.
Disclosure of Invention
The invention aims to provide a method for rapidly detecting dangerous chemicals based on neural network architecture search, so as to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: a dangerous chemical rapid detection method based on neural network architecture search comprises the following steps: acquiring spectral data of dangerous chemicals; step two, designing a network based on the neural network architecture searching method of AutoKeras; step three, obtaining an optimal network architecture;
in the first step, the acquisition of the hazardous chemical spectral data comprises the following steps:
1) acquiring spectral data: measuring spectral data of the samples by using a portable handheld raman instrument, 20 spectra being collected for each sample, for a total of 6000 spectral data;
2) preprocessing spectral data: taking one fourth of 6000 spectral data collected as a test set, namely 1500 spectral data, wherein the test set of each kind is 5 data, and the rest spectral data is taken as a training set and a verification set;
in the second step, the network design of the neural network architecture searching method based on AutoKeras comprises the following steps:
1) constructing a network: the whole network is constructed by adopting a composite block based on batch processing normalization layer, convolution layer, zero filling layer, activation layer and cross-layer connection;
2) extracting effective characteristics: adopting convolution kernels of 1 × 1, 3 × 3 and 1 × 1, carrying out convolution on data according to the convolution kernels, filtering out interference information or unimportant features, and extracting effective features;
3) and (3) relieving the gradient disappearance problem: the gradient disappearance problem is relieved by using the ReLU function, the extraction of the features is accelerated, and the gradient disappearance problem is further relieved by using the training process of the cross-layer connection acceleration network;
in the third step, the obtaining of the optimal network architecture includes the following steps:
1) model training: the AutoKeras searches and integrates network modules, guides preprocessed spectral data into the AutoKeras, utilizes a training set to train a model, utilizes a verification set to check the state and convergence condition of the model, and utilizes a test set to evaluate the generalization capability of the model;
2) performance evaluation and feedback: and automatically optimizing the network architecture until the actual value is smaller than the predicted value to obtain the optimal network architecture.
Preferably, in the step 1), for solid chemicals, the laser is directly irradiated on the surface of the sample at a focal length of 7.5ms to collect the spectral data, and for liquid chemicals, the sample is loaded in a standard sample bottle to collect the spectral data.
Preferably, in the step one 2), the ratio of the number of the spectral data of the training set and the verification set is 4: 1.
Preferably, in the step two 1), the batch normalization layer eliminates negative effects caused by data migration through normalization, so as to alleviate the problem of feature dispersion in the network and make the training of the network more stable and easier.
Preferably, in the step two 1), the whole network additionally uses global average pooling besides the composite block, and the use of global average pooling not only makes the conversion from feature mapping to final classification smoother, but also can improve the over-fitting resistance of the model due to no parameters.
Preferably, in the second step 2), zero values are filled in the upper, lower, left, and right sides of the data, so that the size of the data is always equal to or larger than 3.
Compared with the prior art, the spectrum of dangerous chemicals is measured by the Raman spectrometer, the neural network architecture searching method is introduced, the AutoKeras framework is adopted for developing and designing the self-propelled network architecture, and compared with other neural network methods, the optimal network architecture is found through searching and integrating network modules, performance evaluation and feedback, so that the use threshold of the neural network is reduced, and the time and labor consumption are reduced.
Drawings
FIG. 1 is a flow chart of the steps of the present invention;
FIG. 2 is a flow chart of step three of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1-2, an embodiment of the present invention is shown: a dangerous chemical rapid detection method based on neural network architecture search comprises the following steps: acquiring spectral data of dangerous chemicals; step two, designing a network based on the neural network architecture searching method of AutoKeras; step three, obtaining an optimal network architecture;
in the first step, the acquisition of the hazardous chemical spectral data comprises the following steps:
1) acquiring spectral data: measuring spectral data of samples by using a portable handheld Raman instrument, wherein 20 spectra are collected for each sample, 6000 spectra are collected in total, for solid chemicals, laser is directly irradiated on the surface of the sample at a focal length of 7.5ms to collect the spectral data, and for liquid chemicals, the sample is required to be arranged in a standard sample bottle to collect the spectral data;
2) preprocessing spectral data: taking one fourth of 6000 spectral data collected as a test set, namely 1500 spectral data, wherein each kind of test set is 5 pieces of data, the rest spectral data is taken as a training set and a verification set, and the number ratio of the spectral data of the training set to the spectral data of the verification set is 4: 1;
in the second step, the network design of the neural network architecture searching method based on AutoKeras comprises the following steps:
1) constructing a network: the whole network is constructed by adopting a composite block based on batch processing normalization layer, convolution layer, zero filling layer, activation layer and cross-layer connection; the batch processing normalization layer eliminates negative effects caused by data deviation through normalization, alleviates the problem of characteristic dispersion in the network and enables the training of the network to be more stable and easy; the overall network additionally uses global average pooling besides the composite blocks, the conversion from feature mapping to final classification is smoother due to the use of the global average pooling, and the overfitting resistance of the model can be improved due to the absence of parameters;
2) extracting effective characteristics: adopting convolution kernels of 1 × 1, 3 × 3 and 1 × 1 and carrying out convolution on the data according to the convolution kernels, filtering out interference information or unimportant features, extracting effective features, and filling the upper, lower, left and right sides of the data with zero values to ensure that the size of the data is always more than or equal to 3;
3) and (3) relieving the gradient disappearance problem: the gradient disappearance problem is relieved by using the ReLU function, the extraction of the features is accelerated, and the gradient disappearance problem is further relieved by using the training process of the cross-layer connection acceleration network;
in the third step, the obtaining of the optimal network architecture includes the following steps:
1) model training: the AutoKeras searches and integrates network modules, guides preprocessed spectral data into the AutoKeras, utilizes a training set to train a model, utilizes a verification set to check the state and convergence condition of the model, and utilizes a test set to evaluate the generalization capability of the model;
2) performance evaluation and feedback: and automatically optimizing the network architecture until the actual value is smaller than the predicted value to obtain the optimal network architecture.
Based on the above, the method has the advantages that the method adopts Raman Spectrum (RS) and neural Network Architecture Search (NAS) to detect and identify dangerous chemicals in public safety, and develops a classification model for intelligent identification of the dangerous chemicals based on an AutoKeras framework; the AutoKeras can obtain the results of 100% of ACCC, 100% of ACCcv and 100% of ACCP, namely all chemicals are correctly identified, and the results are obviously superior to the common machine learning methods KNN, RF and SVM and the existing deep networks VGG-16 and DenseNet-BC; meanwhile, the network constructed by AutoKeras has strong feature extraction capability and good robustness; experimental results show that the network generated by AutoKeras is worthy of affirmation, and besides, RS and AutoKeras provide an accurate and robust method for detecting dangerous chemicals, and the method has high application potential in substance detection in other fields such as agriculture and food.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (6)
1. A dangerous chemical rapid detection method based on neural network architecture search comprises the following steps: acquiring spectral data of dangerous chemicals; step two, designing a network based on the neural network architecture searching method of AutoKeras; step three, obtaining an optimal network architecture; the method is characterized in that:
in the first step, the acquisition of the hazardous chemical spectral data comprises the following steps:
1) acquiring spectral data: measuring spectral data of the samples by using a portable handheld raman instrument, 20 spectra being collected for each sample, for a total of 6000 spectral data;
2) preprocessing spectral data: taking one fourth of 6000 spectral data collected as a test set, namely 1500 spectral data, wherein the test set of each kind is 5 data, and the rest spectral data is taken as a training set and a verification set;
in the second step, the network design of the neural network architecture searching method based on AutoKeras comprises the following steps:
1) constructing a network: the whole network is constructed by adopting a composite block based on batch processing normalization layer, convolution layer, zero filling layer, activation layer and cross-layer connection;
2) extracting effective characteristics: adopting convolution kernels of 1 × 1, 3 × 3 and 1 × 1, carrying out convolution on data according to the convolution kernels, filtering out interference information or unimportant features, and extracting effective features;
3) and (3) relieving the gradient disappearance problem: the problem of gradient disappearance is relieved by using a ReLU function, the extraction of features is accelerated, and the problem of gradient disappearance is further relieved by using a cross-layer connection to accelerate the training process of a network;
in the third step, the obtaining of the optimal network architecture includes the following steps:
1) model training: the AutoKeras searches and integrates network modules, guides preprocessed spectral data into the AutoKeras, utilizes a training set to train a model, utilizes a verification set to check the state and convergence condition of the model, and utilizes a test set to evaluate the generalization capability of the model;
2) performance evaluation and feedback: and automatically optimizing the network architecture until the actual value is smaller than the predicted value to obtain the optimal network architecture.
2. The method for rapidly detecting dangerous chemicals based on neural network architecture search according to claim 1, characterized in that: in the step one 1), for solid chemicals, laser is directly irradiated on the surface of a sample at a focal length of 7.5ms to acquire spectral data, and for liquid chemicals, the sample is required to be arranged in a standard sample bottle to acquire the spectral data.
3. The method for rapidly detecting dangerous chemicals based on neural network architecture search according to claim 1, characterized in that: in the step one 2), the quantity ratio of the spectral data of the training set to the spectral data of the verification set is 4: 1.
4. The method for rapidly detecting dangerous chemicals based on neural network architecture search according to claim 1, characterized in that: in the step two 1), the batch processing normalization layer eliminates negative influence caused by data deviation through normalization, alleviates the problem of feature dispersion in the network, and enables the network to be more stable and easier to train.
5. The method for rapidly detecting dangerous chemicals based on neural network architecture search according to claim 1, characterized in that: in the step two 1), the overall network additionally uses global average pooling besides the composite blocks, and the use of the global average pooling not only enables the conversion from feature mapping to final classification to be smoother, but also can improve the overfitting resistance of the model due to the absence of parameters.
6. The method for rapidly detecting dangerous chemicals based on neural network architecture search according to claim 1, characterized in that: in the second step 2), zero values are used for filling up the upper, lower, left and right sides of the data, so that the size of the data is always larger than or equal to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210665879.6A CN115078327A (en) | 2022-06-14 | 2022-06-14 | Rapid detection method for dangerous chemicals based on neural network architecture search |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210665879.6A CN115078327A (en) | 2022-06-14 | 2022-06-14 | Rapid detection method for dangerous chemicals based on neural network architecture search |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115078327A true CN115078327A (en) | 2022-09-20 |
Family
ID=83252261
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210665879.6A Pending CN115078327A (en) | 2022-06-14 | 2022-06-14 | Rapid detection method for dangerous chemicals based on neural network architecture search |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115078327A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116502117A (en) * | 2023-04-13 | 2023-07-28 | 厦门市帕兰提尔科技有限公司 | ResNet-based hazardous chemical identification method, device and equipment |
-
2022
- 2022-06-14 CN CN202210665879.6A patent/CN115078327A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116502117A (en) * | 2023-04-13 | 2023-07-28 | 厦门市帕兰提尔科技有限公司 | ResNet-based hazardous chemical identification method, device and equipment |
| CN116502117B (en) * | 2023-04-13 | 2023-12-15 | 厦门市帕兰提尔科技有限公司 | ResNet-based hazardous chemical identification method, device and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101915744B (en) | Near infrared spectrum nondestructive testing method and device for material component content | |
| CN104596957A (en) | Estimation method for content of copper in soil on basis of visible-light near-infrared spectrum technology | |
| CN101470121A (en) | Built-in bionic smell recognition method and device | |
| CN101995392A (en) | Method for rapidly detecting adulteration of olive oil | |
| CN1831516A (en) | Method for nondistructive discriminating variety and true and false of cigarette using visible light and near-infrared spectrum technology | |
| CN101788517A (en) | Pungent traditional Chinese medicine odor fingerprint map construction system and method based on bionic olfaction | |
| CN103134850A (en) | Tea quality rapid detection apparatus and detection method based on characteristic fragrance | |
| CN103543132B (en) | A kind of coal characteristic measuring method based on wavelet transformation | |
| CN108398416A (en) | A kind of mix ingredients assay method based on laser Raman spectroscopy | |
| CN112345759B (en) | Method for detecting fluorescence intensity peak | |
| Cai et al. | Rapid identification of ore minerals using multi-scale dilated convolutional attention network associated with portable Raman spectroscopy | |
| CN113008805A (en) | Radix angelicae decoction piece quality prediction method based on hyperspectral imaging depth analysis | |
| CN115078327A (en) | Rapid detection method for dangerous chemicals based on neural network architecture search | |
| CN108827909B (en) | Rapid soil classification method based on visible near infrared spectrum and multi-target fusion | |
| CN105223140A (en) | The method for quickly identifying of homology material | |
| CN107247033B (en) | Identify the method for Huanghua Pear maturity based on rapid decay formula life cycle algorithm and PLSDA | |
| CN116858822A (en) | Quantitative analysis method for sulfadiazine in water based on machine learning and Raman spectrum | |
| CN202794093U (en) | Device for fast detecting quality of baked food based on bionic olfaction | |
| CN109668842A (en) | A kind of shellfish heavy metal pollution rapid detection method | |
| CN101140225B (en) | Method for detecting lead in scenting agent with AOTF near-infrared spectrometer | |
| CN110261346A (en) | A method of quickly detecting PVC content in soil based near infrared spectrum | |
| CN105069462A (en) | Qualitative determination method for organic chemical based on spectral characteristic refinement and classifier cascading | |
| CN115420726A (en) | Method for rapidly identifying target object by using reconstructed SERS spectrum | |
| CN108169204A (en) | A kind of Raman spectra pretreatment method based on database | |
| CN107462564A (en) | A kind of aflatoxin B1 rapid free mark detection method based on the super crystal of silver |
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 |