CN107958257A - A kind of Chinese traditional medicinal materials recognition method based on deep neural network - Google Patents
A kind of Chinese traditional medicinal materials recognition method based on deep neural network Download PDFInfo
- Publication number
- CN107958257A CN107958257A CN201710939672.2A CN201710939672A CN107958257A CN 107958257 A CN107958257 A CN 107958257A CN 201710939672 A CN201710939672 A CN 201710939672A CN 107958257 A CN107958257 A CN 107958257A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- training
- chinese medicine
- medicinal materials
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses a kind of Chinese traditional medicinal materials recognition method based on deep neural network, include the following steps:Take pictures web crawlers and manually and gather Chinese medicine picture as the input of data set and pre-processed;Training uses the Bagging methods of integrated study with prediction process, i.e., produces more sub- training sets using stochastical sampling method;Training is finely adjusted to each sub- training set using classical convolutional neural networks model, generates multiple Weak Classifiers, the convolutional neural networks model of use includes AlexNet, SqueezeNet and GoogleNet, finally coordinates Softmax sorting algorithms;And strong classifier is obtained using integrated study combined strategy, classification results are obtained, the integrated study combined strategy uses ballot method.The method of the present invention is used to assist in identifying Chinese medicine, reduces the mistake that amateur occurs in identification, and energy accuracy is high, recognition speed is fast, performance stably analyzes Chinese medicine.
Description
Technical field
The present invention relates to image recognition and integrated study to apply the technical field in Chinese traditional medicinal materials recognition, and in particular to Yi Zhongji
In the Chinese traditional medicinal materials recognition method of deep neural network.
Background technology
Traditional Chinese medicine regards gas, shape, the entity of god as using yin-yang and five elements as theoretical foundation, by human body, by " hoping news ask
Cut " method that ginseng is closed in the four methods of diagnosis, seek the cause of disease, characteristic of disease, sick position, the analysis interpretation of the cause, onset and process of an illness and human five internal organs' six internal organs, channels and collaterals joint, qi-blood-body fluid
Change, judge pathogenic and vital contention, and then draw name of disease, summarize illness type, with dialectical treatmert principle, formulate " sweat, spit, under,
With, temperature, it is clear, mend, disappear " etc. therapy, using a variety for the treatment of means such as Chinese medicine, acupuncture, massage, massage, cup, the qigong, dietotherapy, make
Human body reaches negative and positive and reconciles and rehabilitation.Chinese medicine is one of common and effective treatment means, is used under instruction of Chinese Medicine theory pre-
Anti-, diagnosis, treatment disease or the medicine for adjusting function of human body.Mostly autonomic drug, also have animal drugs, mineral drug and part chemistry,
Biological drug.
Chinese traditional medicinal materials recognition is the required skill of each traditional Chinese medicine practitioner, is also based on China of traditional Chinese medicine traditional culture
The technical ability of the necessary to master of modern healthy service practitioner.From ancient times to the present, the traditional Chinese medical science or traditional Chinese medical science apprentice are from understanding Chinese medicine
Understand its character and effect starts it and is engaged in traditional Chinese medicine career, shennong went into the mountains collecting, tasting and testing different kinds of herbs to be used as medicine, is exactly to be for the first time in human history
Recognize to the scale of system property the process of Chinese herbal medicine and understanding and the study and research of definite effect.
The identification of Chinese medicine at present is expert according to knowledge and experience, or the judgement that picture is compared and made.Although people
To the warm surging pursuit of traditional Chinese medical science health care's idea, but because most of and layman, to knowing in terms of Chinese medicine
Know very deficient;Chinese medicine species is various at the same time, and the good and bad jumbled together in market, or even many procurement staff recognize unclear.In addition, because
Know Chinese medicine by mistake and cause serious consequence also very common.
Image recognition is all an important and popular research direction in computer realm all the time.Deep learning is near
One of most important breakthrough that artificial intelligence field obtains for over ten years, regards in speech recognition, natural language processing, computer
Feel, the numerous areas such as image and video analysis, multimedia all achieves immense success.Now, learnt using deep neural network,
Training pattern, intelligent image identification, so that recognition accuracy is improved, using very extensive.
Chen Yaodan etc. [face identification method [J] Northeast China Normal University's journals (natural science) based on convolutional neural networks,
2016,48(2):70-76.] study and realize a kind of face identification method based on convolutional neural networks;Wang Qian etc. [is based on
Automotive Style Recognition [J] modern computer publications appearing once every ten days of deep neural network, 2015 (12):61-64.] research be based on deep learning
Vehicle automatic identification technology;It is handsome wait [plant leaf blade recognizer research [D] the Beijing Forestry University based on deep learning,
2016.] the deep learning algorithm based on convolutional neural networks is studied;A kind of [Expression Recognition based on deep learning such as Wang Jianyun
Method [J] computers and modernization, 2015 (1):84-87.] propose a kind of expression recognition method based on deep learning.
But due to the particularity of application, recognition of face, Emotion identification, vehicle cab recognition and plant based on deep learning are known
It Deng not apply, it is difficult to be generalized to Chinese traditional medicinal materials recognition.
The content of the invention
The purpose of the present invention is to solve drawbacks described above of the prior art, there is provided a kind of based on deep neural network
Chinese traditional medicinal materials recognition method, deep learning is combined with Chinese traditional medicinal materials recognition, is carried out using classical convolutional neural networks Models Sets micro-
Training is adjusted, then recognition accuracy is improved using Ensemble Learning Algorithms, for assisting in identifying Chinese medicine, amateur is reduced and is identifying
The mistake of middle appearance, and can accuracy is high, recognition speed is fast, performance stably analyzes Chinese medicine.
The purpose of the present invention can be reached by adopting the following technical scheme that:
A kind of Chinese traditional medicinal materials recognition method based on deep neural network, the Chinese traditional medicinal materials recognition method include following step
Suddenly:
The input of S1, collection Chinese medicine picture as data set, the Chinese medicine picture of collection is pre-processed;
S2, using stochastical sampling method produce multiple sub- training sets of Chinese medicine;
S3, using convolutional neural networks model be finely adjusted training to the sub- training set of each Chinese medicine, generates multiple Chinese medicines
Material identifies Weak Classifier;
S4, using integrated study combined strategy obtain Chinese traditional medicinal materials recognition strong classifier, obtains Classification of Chinese Drug result.
Further, Chinese medicine picture is gathered by web crawlers and artificial photographic method in the step S1 and is used as number
According to the input of collection, then the unbalanced Chinese medicine classification of quantity is equalized.
Further, the entirety training in the step S2, step S3, step S4 uses integrated study with prediction process
Bagging methods.
Further, stochastical sampling method uses self-service sampling method in the step S2, is obtained by self-service sampling method
Multiple variant sub- training sets.
Further, by introducing deep learning technology in the step S3, using convolutional neural networks model to each
A sub- training set of Chinese medicine is finely adjusted training, generates multiple Chinese traditional medicinal materials recognition Weak Classifiers, wherein, the convolutional Neural net
Network model includes AlexNet, SqueezeNet and GoogleNet.
Further, the step S3 Chinese traditional medicinal materials recognitions Weak Classifier using convolutional neural networks make feature extraction and
Make to classify using Softmax graders, it is specific as follows:
For T sub- training sets, training is finely adjusted to each sub- training set using classical convolutional neural networks model,
Generate T Weak Classifier;
Chinese medicine picture is as inputting, by multiple convolutional layers and down-sampling layer output characteristic figure, wherein, the public affairs of convolutional layer
Formula is as follows:
Conv_output=f (WTx+b)
Wherein f (*) represents activation primitive, and x represents input data, W and b representation parameters, the present invention is using ReLU activation letters
Number, formula are as follows:
F (x)=max (0, x)
Its derivation form is:
The formula of down-sampling layer is as follows, using Max-pooling methods:
Pooling_output=max (X)
Wherein X represents n*n matrixes;
The input of full articulamentum is a vector, and obtaining vector by rasterisation is connected to full articulamentum, finally by
Softmax graders obtain the classification results of the grader, which receives the input data of 4096 dimensions, output
98 dimensions as a result, the result represents the confidence level that the input sample corresponds to 98 Chinese medicine classifications, where then taking its maximum
Classification is classification results, and the formula of Softmax graders is as follows:
Wherein j=1,2 ..., K, K represent the number of class, wherein value 98, z=WT x+b, W, b be Softmax parameter, x
For the input feature vector of 4096 dimensions.
Further, training algorithm uses stochastic gradient descent method in the step S3, and training connects this method every time
By a certain number of training datas, by after, the difference of output and data physical tags is weighed using loss function before network
Away from, carry out reverse train network parameter followed by this measurement, wherein, loss function uses cross entropy loss function and cohesion
The combination of loss function, formula are as follows:
Wherein, λ is used for controlling ratio of the cohesion loss function in total loss function, xiRepresent original signal, ziGeneration
Table reconstruction signal, represents that length is d in the form of vectors, and can be transform as the form of inner product of vectors easily, and K represents every
The sample number of secondary iteration, CyiRepresent yiThe class heart of category feature, yiFor original signal xiClass label.
Further, integrated study combined strategy uses ballot method in the step S4.
The present invention is had the following advantages relative to the prior art and effect:
1st, the method for the present invention utilizes the self-service sampling method in stochastical sampling, puts back to more sub-samplings, produces multiple variant
The sub- training set of Chinese medicine of property;
2nd, the method for the present invention uses convolutional neural networks, automatically learns with strongly expressed power by convolutional neural networks
Convolution kernel and these convolution kernels combination, obtain best Chinese medicine feature representation;
3rd, the method for the present invention uses Softmax sorting techniques, is a kind of supervised learning method, suitable for more classification problems,
Classifying quality is notable;
4th, the method for the present invention uses the Bagging methods of integrated study, and classification accuracy is apparently higher than single grader;
5th, the method for the present invention uses and is combined deep learning with Chinese traditional medicinal materials recognition, compared with conventional method, accuracy height,
Recognition speed is fast, performance is stablized, and this method has certain market value and promotional value.
Brief description of the drawings
Fig. 1 is a kind of step flow chart of Chinese traditional medicinal materials recognition method based on deep neural network disclosed in the present invention;
Fig. 2 be in a kind of Chinese traditional medicinal materials recognition method based on deep neural network disclosed in the present invention integrally training with it is pre-
The schematic diagram of the Bagging methods of integrated study during survey;
Fig. 3 is convolutional neural networks in a kind of Chinese traditional medicinal materials recognition method based on deep neural network disclosed in the present invention
General structure schematic diagram.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical solution in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
Part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
All other embodiments obtained without making creative work, belong to the scope of protection of the invention.
Embodiment
A kind of Chinese traditional medicinal materials recognition method based on deep neural network, includes the following steps:
The input of S1, collection Chinese medicine picture as data set, the Chinese medicine picture of collection is pre-processed;
In above-mentioned steps S1, input of the collection Chinese medicine picture as data set, wherein Chinese medicine image data collection is logical
Web crawlers and artificial photographic method are crossed, and the image data set marked by artificial screening, size are normalized to 256*256,
And needed data set being divided into training set and test set, ratio 7 according to neural network algorithm:2.Due to partial category Chinese medicine
Material picture is less, causes whole data set quantitatively unbalanced, occurs over-fitting situation in the training process in order to prevent, right
Training set uses clone method, allows each classification quantitatively to reach consistent;Also, considered based on neural network properties, to defeated
Enter data and carry out equalization operation, formula is as follows:
Wherein, mean value calculation formula is:
R, G, B value is respectively the rgb value in pixel, Rmij, Gmij, Bmij be m-th picture sample the (i,
J) rgb value of a pixel.
S2, using stochastical sampling method produce multiple sub- training sets of Chinese medicine;
S3, be finely adjusted each Chinese medicine training set using classical convolutional neural networks model training, and generation is more
A Chinese traditional medicinal materials recognition Weak Classifier;
S4, using integrated study combined strategy obtain Chinese traditional medicinal materials recognition strong classifier, obtains Classification of Chinese Drug result.
In above-mentioned steps S2, S3, S4, overall training uses the Bagging methods of integrated study, the process with prediction process
As shown in Figure 2.
Integrated study is to be combined into a strong classifier with multiple or multiple Weak Classifiers, so as to reach lifting sorting technique
Effect, it is instrument very powerful in machine learning algorithm, its thought is simple, by gathering the ability of multiple models, effect
Fruit exceedes the ability of single model, and main method has two methods of Bagging and Boosting.Bagging is used in the present invention
Method, the main distinction in Bagging methods Weak Classifier be that parallelization produces, and Boosting methods are to be based on weights,
The generation of Weak Classifier is to pass through serial mode.
Bagging methods, flow are summarized as follows:
First, by statistics stochastical sampling method, used here as self-service sampling method, multiple variant son instructions are obtained
Practice collection.Self-service sampling method is using there is the mode for putting back to repeated sampling to carry out data acquisition, i.e., from training set size is every time m's
In data set, take a sample then the sample to be put back to as an element in sub- training set, repeat behavior m times, this
Sample has just obtained the sub- training set that size is m, some samples repeat in sub- training set, some samples do not occur then
Cross.The above process, which is repeated several times, can obtain multiple sub- training sets with otherness.The reason for carrying out this method sampling:
The probability that each sample is not adopted is:
M sampling is carried out, the probability which is not adopted is:
So the limit is taken to have
It is approximately equal to 0.368, therefore about 36.8% data do not occur in sub- training set on original training set
Cross.
Next, introduce deep learning technology, using classical convolutional neural networks model include AlexNet,
SqueezeNet and GoogleNet, is finely adjusted training generation Chinese traditional medicinal materials recognition Weak Classifier;
The advantage of convolutional neural networks is to set up hierarchical structure, is obtained automatically with reference to the training method of mass data
Must have the feature of strongly expressed power, while its internal using weights shared mechanism, reduce training parameter, make the network of complexity
Structure becomes simpler and is easy to trained.In the convolutional neural networks used in the present invention, AlexNet is volume the most classical
Product neutral net, including 5 convolution and down-sampling layer, 3 full articulamentum compositions;SqueezeNet is primarily to reduce CNN
Model parameter quantity and design, parameter reduces 50 times on Imagenet data sets, and model size is substantially reduced;
The GoogleNet network numbers of plies have 22 layers, and expression effect is optimal in 3 kinds of selected networks.
It is finely adjusted and is trained for the weak of Chinese medicine data set using the model and Chinese medicine data set of these three networks
Grader.Sorting algorithm uses Softmax classifier algorithms, and this method is suitable for more classification problems, at the same can directly with volume
Product neutral net is connected, and the classifier algorithm can be led everywhere, therefore can realize training method end to end.Specifically such as
Under:
For T sub- training sets, training is finely adjusted to each sub- training set using classical convolutional neural networks model,
Generate T Weak Classifier.
Be illustrated in figure 3 the general structure of convolutional neural networks, Chinese medicine picture as input, by multiple convolutional layers and
Down-sampling layer output characteristic figure, wherein, the formula of convolutional layer is as follows:
Conv_output=f (WTx+b)
Wherein f (*) represents activation primitive, and x represents input data, W and b representation parameters, the present invention is using ReLU activation letters
Number, formula are as follows:
F (x)=max (0, x)
Its derivation form is:
The formula of down-sampling layer is as follows, and the present invention uses Max-pooling methods:
Pooling_output=max (X)
Wherein X represents n*n matrixes.
But the input of full articulamentum is a vector, therefore vector is obtained by rasterisation and is connected to full articulamentum, most
The classification results of the grader are obtained by Softmax graders afterwards.
The Softmax graders receive 4096 dimension input datas, output 98 dimension as a result, the result represents the input sample
The confidence level of 98 Chinese medicine classifications of this correspondence, it is classification results then to take classification where its maximum, and the formula of Softmax is such as
Under:
Wherein j=1,2 ..., K, K represent the number of class, K=98 in the present invention.
Wherein z=WT x+b, W, b are the parameter of Softmax, and x is the input feature vector of 4096 dimensions.
The training algorithm of this method uses stochastic gradient descent method, and training receives a certain number of training to this method every time
Data, by after, the gap of output and data physical tags is weighed using loss function before network, followed by this weighing apparatus
Amount carrys out reverse train network parameter, wherein, loss function employs the combination of cross entropy loss function and cohesion loss function
Body, formula are as follows:
A. cross entropy loss function:
Wherein x represents original signal, and z represents reconstruction signal, in the form of vectors represent length be d, and can easily by
It transform the form of inner product of vectors as, and K represents the sample number of each iteration.
B. cohesion loss function:
Wherein, K represents the sample number of each iteration, CyiRepresent yiThe class heart of category feature, yiFor original signal xiClassification
Label.
We use the combination of both loss functions, as follows:
Wherein, λ is used for controlling ratio of the cohesion loss function in total loss function.
Finally, strong classifier is obtained using integrated study combined strategy, so as to obtain classification results, wherein combined strategy is adopted
With simple vote method:Assuming that prediction classification is { c1,c2,...cK, for any one forecast sample x, T Weak Classifier it is pre-
It is (h respectively to survey result1(x),h2(x)...hT(x)).Simple vote method is that the minority is subordinate to the majority, that is, T Weak Classifier
To in the prediction result of sample x, the most classification c of quantityiFor final class categories, if more than one classification obtains highest
Ticket, then randomly choose one and do final classification.
Embodiment two
A kind of base is specifically introduced in terms of the present embodiment is built from frame, data set preparation, model training, actual test are several
In the method for the Chinese traditional medicinal materials recognition of deep neural network, detailed process is introduced as follows.
First, frame build process is as follows:
The 1st, GPU drivings and computing environment are installed;
2nd, fitting deep learning framework Caffe environment.
2nd, data set set-up procedure is as follows:
1st, web crawlers instrument for Chinese medicine image data is write using Python, using the instrument multi-thread
Network Chinese medicine picture is gathered in the case of journey and records Chinese medicine label, and the Chinese medicine picture to collecting automatically, is carried out just
Walk artificial screening;
2nd, video capture is carried out to Chinese medicine using high-definition camera to medicinal material shop, and using keyframe techniques extraction wherein
Some pictures as Chinese medicine supplementary data;
3rd, the data that above two mode obtains are integrated;
4th, for categorical measure imbalance problem, using random reproduction method by its equilibrating;
5th, equalization processing is carried out to Chinese medicine image data;
6th, the Chinese medicine picture being disposed is converted into the receivable lmdb data lattice of Caffe frames together with its label
Formula.
3rd, training process is as follows:
1st, AlexNet, SqueezeNet and GoogleNet convolutional Neural are downloaded by Caffe public networks distribution platform
Network model;
2nd, the output node number of complete last layer of articulamentum of above-mentioned network is changed to 98, because the Chinese medicine data set of collection
Classification number is 98 classes;
3rd, the lmdb formatted datas for obtaining data set set-up procedure input convolutional neural networks;
4th, by multiple convolutional layers in AlexNet or SqueezeNet or GoogleNet convolutional neural networks models and
Down-sampling layer output characteristic figure;
5th, rasterization process obtains a feature vector and is connected to full articulamentum;
6th, by Softmax graders obtain the grader as a result, sending the result to loss function layer;
7th, according to result and cross entropy-cohesion loss function, its loss and passback gradient are calculated;
8th, by back-propagation algorithm, convolutional neural networks parameter is adjusted;Wherein back-propagation algorithm is using following super ginseng
Number, which obtained by multiple cross validation:
Parameter name | Value |
batch_size | 64 |
base_lr | 0.001 |
max_iter | 30000 |
9th, the process of above 3-7 is repeated, until loss function value is less than threshold value or maximum iteration, training terminates.
4th, test process is as follows:
Multithreading test script is write using Python, including following operation:
1st, Chinese medicine picture size to be tested is normalized to 256*256, and completes equalization operation;
2nd, be loaded into the obtained each Weak Classifier of training, i.e., after fine setting, AlexNet suitable for Chinese medicine,
SqueezeNet and GoogleNet convolutional neural networks models;
3rd, by the picture to be tested after processing in a manner of blockette, the picture of 10 224*224 sizes is obtained;
4th, 10 pictures are all inputted into all Weak Classifiers by more than, obtain the recognition result of each weak typing;
5th, the ballot mode of each one ticket of Weak Classifier, the classification for obtaining most polls are the classification knots of final strong classifier
Fruit, that is, the Chinese medicine classification predicted.
In conclusion the method for the present invention utilizes the self-service sampling method in stochastical sampling, more sub-samplings are put back to, are produced multiple
The sub- training set of variant Chinese medicine;Using convolutional neural networks, automatically learn to have by convolutional neural networks strong
The combination of the convolution kernel of expressiveness and these convolution kernels, obtains best Chinese medicine feature representation;Using Softmax points
Class method, is a kind of supervised learning method, notable suitable for more classification problems, classifying quality;Using the Bagging of integrated study
Method, classification accuracy is apparently higher than single grader;It is combined using by deep learning with Chinese traditional medicinal materials recognition, for aiding in knowing
Other Chinese medicine, reduces the mistake that amateur occurs in identification, and compared with conventional method, can accuracy height, recognition speed
Hurry up, performance stably analyzes Chinese medicine, this method has certain market value and promotional value.
Above-described embodiment is the preferable embodiment of the present invention, but embodiments of the present invention and from above-described embodiment
Limitation, other any Spirit Essences without departing from the present invention with made under principle change, modification, replacement, combine, simplification,
Equivalent substitute mode is should be, is included within protection scope of the present invention.
Claims (8)
- A kind of 1. Chinese traditional medicinal materials recognition method based on deep neural network, it is characterised in that the Chinese traditional medicinal materials recognition method bag Include following steps:The input of S1, collection Chinese medicine picture as data set, the Chinese medicine picture of collection is pre-processed;S2, using stochastical sampling method produce multiple sub- training sets of Chinese medicine;S3, using convolutional neural networks model be finely adjusted training to the sub- training set of each Chinese medicine, generates multiple Chinese medicines and knows Other Weak Classifier;S4, using integrated study combined strategy obtain Chinese traditional medicinal materials recognition strong classifier, obtains Classification of Chinese Drug result.
- 2. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 1, it is characterised in that described Step S1 in the input that Chinese medicine picture is used as data set gathered by web crawlers and artificial photographic method, then to quantity Unbalanced Chinese medicine classification equalization.
- 3. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 1, it is characterised in that described Step S2, step S3, the entirety training in step S4 the Bagging methods of integrated study are used with prediction process.
- 4. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 1, it is characterised in that described Step S2 in stochastical sampling method use self-service sampling method, pass through self-service sampling method and obtain multiple variant son training Collection.
- 5. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 1, it is characterised in that described Step S3 in by introducing deep learning technology, using convolutional neural networks model to the sub- training set of each Chinese medicine carry out it is micro- Adjust training, generate multiple Chinese traditional medicinal materials recognition Weak Classifiers, wherein, the convolutional neural networks model include AlexNet, SqueezeNet and GoogleNet.
- 6. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 5, it is characterised in that described Step S3 Chinese traditional medicinal materials recognitions Weak Classifier make feature extraction using convolutional neural networks and divided using Softmax graders Class, it is specific as follows:For T sub- training sets, training is finely adjusted to each sub- training set using classical convolutional neural networks model, is generated T Weak Classifier;Chinese medicine picture is as inputting, by multiple convolutional layers and down-sampling layer output characteristic figure, wherein, the formula of convolutional layer is such as Under:Conv_output=f (WTx+b)Wherein f (*) represents activation primitive, and x represents input data, W and b representation parameters, and the present invention uses ReLU activation primitives, public Formula is as follows:F (x)=max (0, x)Its derivation form is:<mrow> <msup> <mi>f</mi> <mo>&prime;</mo> </msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mn>1</mn> <mo>,</mo> <mi>x</mi> <mo>></mo> <mn>0</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mn>0</mn> <mo>,</mo> <mi>x</mi> <mo>&le;</mo> <mn>0</mn> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>The formula of down-sampling layer is as follows, using Max-pooling methods:Pooling_output=max (X)Wherein X represents n*n matrixes;The input of full articulamentum is a vector, and obtaining vector by rasterisation is connected to full articulamentum, finally by Softmax Grader obtains the classification results of the grader, which receives the input data of 4096 dimensions, the knot of the dimension of output 98 Fruit, the result represent the confidence level that the input sample corresponds to 98 Chinese medicine classifications, and classification where then taking its maximum is point Class is as a result, the formula of Softmax graders is as follows:<mrow> <mi>f</mi> <msub> <mrow> <mo>(</mo> <mi>z</mi> <mo>)</mo> </mrow> <mi>j</mi> </msub> <mo>=</mo> <mfrac> <msup> <mi>e</mi> <msub> <mi>z</mi> <mi>j</mi> </msub> </msup> <mrow> <munderover> <mo>&Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </munderover> <msup> <mi>e</mi> <msub> <mi>z</mi> <mi>k</mi> </msub> </msup> </mrow> </mfrac> </mrow>Wherein j=1,2 ..., K, K represent the number of class, wherein value 98, z=WTX+b, W, b are the parameter of Softmax, and x is The input feature vector of 4096 dimensions.
- 7. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 5, it is characterised in that described Step S3 in training algorithm use stochastic gradient descent method, training receives a certain number of training datas to this method every time, By after, the gap of output and data physical tags is weighed using loss function before network, come followed by this measurement Reverse train network parameter, wherein, loss function is using the combination of cross entropy loss function and cohesion loss function, formula It is as follows:<mrow> <mi>L</mi> <mo>=</mo> <msub> <mi>L</mi> <mi>H</mi> </msub> <mo>+</mo> <msub> <mi>&lambda;L</mi> <mi>C</mi> </msub> <mo>=</mo> <mo>-</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </munderover> <msub> <mi>x</mi> <mi>i</mi> </msub> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <mfrac> <mi>&lambda;</mi> <mn>2</mn> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </munderover> <mo>|</mo> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>c</mi> <msub> <mi>y</mi> <mi>i</mi> </msub> </msub> <mo>|</mo> <msubsup> <mo>|</mo> <mn>2</mn> <mn>2</mn> </msubsup> </mrow>Wherein, λ is used for controlling ratio of the cohesion loss function in total loss function, xiRepresent original signal, ziRepresent weight Structure signal, represents that length is d in the form of vectors, and can be transform as the form of inner product of vectors easily, and K is represented every time repeatedly The sample number in generation, CyiRepresent yiThe class heart of category feature, yiFor original signal xiClass label.
- 8. a kind of Chinese traditional medicinal materials recognition method based on deep neural network according to claim 1, it is characterised in that described Step S4 in integrated study combined strategy use ballot method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710939672.2A CN107958257A (en) | 2017-10-11 | 2017-10-11 | A kind of Chinese traditional medicinal materials recognition method based on deep neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710939672.2A CN107958257A (en) | 2017-10-11 | 2017-10-11 | A kind of Chinese traditional medicinal materials recognition method based on deep neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107958257A true CN107958257A (en) | 2018-04-24 |
Family
ID=61953981
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710939672.2A Pending CN107958257A (en) | 2017-10-11 | 2017-10-11 | A kind of Chinese traditional medicinal materials recognition method based on deep neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107958257A (en) |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108776805A (en) * | 2018-05-03 | 2018-11-09 | 北斗导航位置服务(北京)有限公司 | It is a kind of establish image classification model, characteristics of image classification method and device |
CN108830254A (en) * | 2018-06-27 | 2018-11-16 | 福州大学 | A kind of detection of fine granularity vehicle and recognition methods based on data balancing strategy and intensive attention network |
CN109142374A (en) * | 2018-08-15 | 2019-01-04 | 广州市心鉴智控科技有限公司 | Method and system based on the efficient Checking model of extra small sample training |
CN109190643A (en) * | 2018-09-14 | 2019-01-11 | 华东交通大学 | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment |
CN109276255A (en) * | 2018-11-27 | 2019-01-29 | 平安科技(深圳)有限公司 | A kind of limb tremor detection method and device |
CN109522924A (en) * | 2018-09-28 | 2019-03-26 | 浙江农林大学 | A kind of broad-leaf forest wood recognition method based on single photo |
CN109615010A (en) * | 2018-12-13 | 2019-04-12 | 济南大学 | Chinese traditional medicinal materials recognition method and system based on double scale convolutional neural networks |
CN109614994A (en) * | 2018-11-27 | 2019-04-12 | 佛山市奥策科技有限公司 | A kind of tile typology recognition methods and device |
CN109857861A (en) * | 2019-01-04 | 2019-06-07 | 平安科技(深圳)有限公司 | File classification method, device, server and medium based on convolutional neural networks |
CN109919224A (en) * | 2019-03-06 | 2019-06-21 | 颐保医疗科技(上海)有限公司 | A kind of discrimination method of the Chinese medicine that interweaves truth with fiction based on artificial intelligence |
CN109948676A (en) * | 2019-03-06 | 2019-06-28 | 颐保医疗科技(上海)有限公司 | A kind of discrimination method in the Chinese medicine plantation place of production based on artificial intelligence |
CN109948805A (en) * | 2019-03-06 | 2019-06-28 | 颐保医疗科技(上海)有限公司 | It is a kind of applied to the artificial intelligence identifying system of pharmaceutical factory and its recognition methods |
CN109978018A (en) * | 2019-03-06 | 2019-07-05 | 颐保医疗科技(上海)有限公司 | A kind of Chinese medicine artificial intelligence identification system and its discrimination method |
CN110110728A (en) * | 2019-05-16 | 2019-08-09 | 湖南睿图智能科技有限公司 | A kind of Chinese traditional medicinal materials recognition method |
CN110135261A (en) * | 2019-04-15 | 2019-08-16 | 北京易华录信息技术股份有限公司 | A kind of method and system of trained road anomalous identification model, road anomalous identification |
CN110164519A (en) * | 2019-05-06 | 2019-08-23 | 北京工业大学 | A kind of classification method for being used to handle electronic health record blended data based on many intelligence networks |
CN110414548A (en) * | 2019-06-06 | 2019-11-05 | 西安电子科技大学 | The level Bagging method of sentiment analysis is carried out based on EEG signals |
CN110610136A (en) * | 2019-08-21 | 2019-12-24 | 广西电网有限责任公司电力科学研究院 | Transformer substation equipment identification module and identification method based on deep learning |
CN110826683A (en) * | 2018-08-07 | 2020-02-21 | 北京京东尚科信息技术有限公司 | Method and device for predicting user behavior occurrence probability |
CN110850020A (en) * | 2019-11-11 | 2020-02-28 | 中国药科大学 | Traditional Chinese medicine identification method based on artificial intelligence |
CN111191782A (en) * | 2018-11-14 | 2020-05-22 | 普天信息技术有限公司 | Convolutional network training method and device |
CN111460913A (en) * | 2020-03-13 | 2020-07-28 | 北京理工大学 | Self-adaptive time sequence signal directivity prediction method |
CN111582328A (en) * | 2020-04-22 | 2020-08-25 | 北方民族大学 | COVID-19 classification identification method based on lung CT image |
CN111598130A (en) * | 2020-04-08 | 2020-08-28 | 天津大学 | Traditional Chinese medicine identification method based on multi-view convolutional neural network |
CN111797853A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Feature extraction method and device, storage medium and electronic equipment |
CN112495841A (en) * | 2021-02-04 | 2021-03-16 | 中国科学院自动化研究所 | Traditional Chinese medicine sorting system based on neural network |
CN113610184A (en) * | 2021-08-19 | 2021-11-05 | 江西应用技术职业学院 | Wood texture classification method based on transfer learning |
CN114549939A (en) * | 2022-04-26 | 2022-05-27 | 汉谷云智(武汉)科技有限公司 | Tree pest and disease picture classification method based on image feature extraction |
CN114998639B (en) * | 2022-04-19 | 2024-04-26 | 安徽农业大学 | Deep learning-based traditional Chinese medicine category identification method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160180195A1 (en) * | 2013-09-06 | 2016-06-23 | Toyota Jidosha Kabushiki Kaisha | Augmenting Layer-Based Object Detection With Deep Convolutional Neural Networks |
CN106951825A (en) * | 2017-02-13 | 2017-07-14 | 北京飞搜科技有限公司 | A kind of quality of human face image assessment system and implementation method |
CN107092906A (en) * | 2017-05-01 | 2017-08-25 | 刘至键 | A kind of Chinese traditional medicinal materials recognition device based on deep learning |
CN107103338A (en) * | 2017-05-19 | 2017-08-29 | 杭州电子科技大学 | Merge the SAR target identification methods of convolution feature and the integrated learning machine that transfinites |
-
2017
- 2017-10-11 CN CN201710939672.2A patent/CN107958257A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160180195A1 (en) * | 2013-09-06 | 2016-06-23 | Toyota Jidosha Kabushiki Kaisha | Augmenting Layer-Based Object Detection With Deep Convolutional Neural Networks |
CN106951825A (en) * | 2017-02-13 | 2017-07-14 | 北京飞搜科技有限公司 | A kind of quality of human face image assessment system and implementation method |
CN107092906A (en) * | 2017-05-01 | 2017-08-25 | 刘至键 | A kind of Chinese traditional medicinal materials recognition device based on deep learning |
CN107103338A (en) * | 2017-05-19 | 2017-08-29 | 杭州电子科技大学 | Merge the SAR target identification methods of convolution feature and the integrated learning machine that transfinites |
Non-Patent Citations (4)
Title |
---|
SHIMENG YANG等: "A Light and Discriminative Deep Networks for Off-line Handwritten Chinese Character Recognition", 《2017 32ND YOUTH ACADEMIC ANNUAL CONFERENCE OF CHINESE ASSOCIATION OF AUTOMATION》 * |
刘余霞等: "基于Bagging集成学习的字符识别方法", 《计算机工程与应用》 * |
孙鑫等: "基于深度卷积网络的中药饮片图像识别", 《世界科学技术-中医药现代化》 * |
王化祥: "《现代传技术及应用》", 30 June 2016, 天津:天津大学出版社 * |
Cited By (36)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108776805A (en) * | 2018-05-03 | 2018-11-09 | 北斗导航位置服务(北京)有限公司 | It is a kind of establish image classification model, characteristics of image classification method and device |
CN108830254A (en) * | 2018-06-27 | 2018-11-16 | 福州大学 | A kind of detection of fine granularity vehicle and recognition methods based on data balancing strategy and intensive attention network |
CN108830254B (en) * | 2018-06-27 | 2021-10-29 | 福州大学 | Fine-grained vehicle type detection and identification method based on data balance strategy and intensive attention network |
CN110826683A (en) * | 2018-08-07 | 2020-02-21 | 北京京东尚科信息技术有限公司 | Method and device for predicting user behavior occurrence probability |
CN109142374A (en) * | 2018-08-15 | 2019-01-04 | 广州市心鉴智控科技有限公司 | Method and system based on the efficient Checking model of extra small sample training |
CN109190643A (en) * | 2018-09-14 | 2019-01-11 | 华东交通大学 | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment |
CN109522924A (en) * | 2018-09-28 | 2019-03-26 | 浙江农林大学 | A kind of broad-leaf forest wood recognition method based on single photo |
CN111191782A (en) * | 2018-11-14 | 2020-05-22 | 普天信息技术有限公司 | Convolutional network training method and device |
CN109614994A (en) * | 2018-11-27 | 2019-04-12 | 佛山市奥策科技有限公司 | A kind of tile typology recognition methods and device |
CN109276255B (en) * | 2018-11-27 | 2023-02-28 | 平安科技(深圳)有限公司 | Method and device for detecting tremor of limbs |
CN109276255A (en) * | 2018-11-27 | 2019-01-29 | 平安科技(深圳)有限公司 | A kind of limb tremor detection method and device |
CN109615010A (en) * | 2018-12-13 | 2019-04-12 | 济南大学 | Chinese traditional medicinal materials recognition method and system based on double scale convolutional neural networks |
CN109615010B (en) * | 2018-12-13 | 2020-11-10 | 济南大学 | Traditional Chinese medicine material identification method and system based on double-scale convolutional neural network |
CN109857861A (en) * | 2019-01-04 | 2019-06-07 | 平安科技(深圳)有限公司 | File classification method, device, server and medium based on convolutional neural networks |
CN109919224A (en) * | 2019-03-06 | 2019-06-21 | 颐保医疗科技(上海)有限公司 | A kind of discrimination method of the Chinese medicine that interweaves truth with fiction based on artificial intelligence |
CN109948676A (en) * | 2019-03-06 | 2019-06-28 | 颐保医疗科技(上海)有限公司 | A kind of discrimination method in the Chinese medicine plantation place of production based on artificial intelligence |
CN109948805A (en) * | 2019-03-06 | 2019-06-28 | 颐保医疗科技(上海)有限公司 | It is a kind of applied to the artificial intelligence identifying system of pharmaceutical factory and its recognition methods |
CN109978018A (en) * | 2019-03-06 | 2019-07-05 | 颐保医疗科技(上海)有限公司 | A kind of Chinese medicine artificial intelligence identification system and its discrimination method |
CN111797853A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Feature extraction method and device, storage medium and electronic equipment |
CN110135261A (en) * | 2019-04-15 | 2019-08-16 | 北京易华录信息技术股份有限公司 | A kind of method and system of trained road anomalous identification model, road anomalous identification |
CN110164519A (en) * | 2019-05-06 | 2019-08-23 | 北京工业大学 | A kind of classification method for being used to handle electronic health record blended data based on many intelligence networks |
CN110110728A (en) * | 2019-05-16 | 2019-08-09 | 湖南睿图智能科技有限公司 | A kind of Chinese traditional medicinal materials recognition method |
CN110414548A (en) * | 2019-06-06 | 2019-11-05 | 西安电子科技大学 | The level Bagging method of sentiment analysis is carried out based on EEG signals |
CN110610136A (en) * | 2019-08-21 | 2019-12-24 | 广西电网有限责任公司电力科学研究院 | Transformer substation equipment identification module and identification method based on deep learning |
CN110610136B (en) * | 2019-08-21 | 2023-07-18 | 广西电网有限责任公司电力科学研究院 | Substation equipment identification module and method based on deep learning |
CN110850020A (en) * | 2019-11-11 | 2020-02-28 | 中国药科大学 | Traditional Chinese medicine identification method based on artificial intelligence |
CN110850020B (en) * | 2019-11-11 | 2022-03-29 | 中国药科大学 | Traditional Chinese medicine identification method based on artificial intelligence |
CN111460913A (en) * | 2020-03-13 | 2020-07-28 | 北京理工大学 | Self-adaptive time sequence signal directivity prediction method |
CN111598130A (en) * | 2020-04-08 | 2020-08-28 | 天津大学 | Traditional Chinese medicine identification method based on multi-view convolutional neural network |
CN111582328A (en) * | 2020-04-22 | 2020-08-25 | 北方民族大学 | COVID-19 classification identification method based on lung CT image |
CN112495841A (en) * | 2021-02-04 | 2021-03-16 | 中国科学院自动化研究所 | Traditional Chinese medicine sorting system based on neural network |
CN112495841B (en) * | 2021-02-04 | 2021-05-11 | 中国科学院自动化研究所 | Traditional Chinese medicine sorting system based on neural network |
CN113610184A (en) * | 2021-08-19 | 2021-11-05 | 江西应用技术职业学院 | Wood texture classification method based on transfer learning |
CN114998639B (en) * | 2022-04-19 | 2024-04-26 | 安徽农业大学 | Deep learning-based traditional Chinese medicine category identification method |
CN114549939A (en) * | 2022-04-26 | 2022-05-27 | 汉谷云智(武汉)科技有限公司 | Tree pest and disease picture classification method based on image feature extraction |
CN114549939B (en) * | 2022-04-26 | 2022-09-23 | 汉谷云智(武汉)科技有限公司 | Tree disease and pest image classification method based on image feature extraction |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107958257A (en) | A kind of Chinese traditional medicinal materials recognition method based on deep neural network | |
CN106682616A (en) | Newborn-painful-expression recognition method based on dual-channel-characteristic deep learning | |
CN109522973A (en) | Medical big data classification method and system based on production confrontation network and semi-supervised learning | |
CN103996056B (en) | Tattoo image classification method based on deep learning | |
CN109949257A (en) | Area-of-interest compressed sensing image reconstructing method based on deep learning | |
Rajayogi et al. | Indian food image classification with transfer learning | |
CN107016681A (en) | Brain MRI lesion segmentation approach based on full convolutional network | |
CN110069958A (en) | A kind of EEG signals method for quickly identifying of dense depth convolutional neural networks | |
CN107016406A (en) | The pest and disease damage image generating method of network is resisted based on production | |
CN110197729A (en) | Tranquillization state fMRI data classification method and device based on deep learning | |
CN108764128A (en) | A kind of video actions recognition methods based on sparse time slice network | |
CN108491849A (en) | Hyperspectral image classification method based on three-dimensional dense connection convolutional neural networks | |
CN106485259B (en) | A kind of image classification method based on high constraint high dispersive principal component analysis network | |
CN109508644A (en) | Facial paralysis grade assessment system based on the analysis of deep video data | |
CN107424145A (en) | The dividing method of nuclear magnetic resonance image based on three-dimensional full convolutional neural networks | |
CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
Zhou et al. | Multi-classification of skin diseases for dermoscopy images using deep learning | |
CN109902736A (en) | A kind of Lung neoplasm image classification method indicated based on autocoder construction feature | |
CN107194426A (en) | A kind of image-recognizing method based on Spiking neutral nets | |
CN109389171A (en) | Medical image classification method based on more granularity convolution noise reduction autocoder technologies | |
CN104484886B (en) | A kind of dividing method and device of MR images | |
CN106909938A (en) | Viewing angle independence Activity recognition method based on deep learning network | |
CN112472048B (en) | Method for realizing neural network for identifying pulse condition of cardiovascular disease patient | |
CN110399821A (en) | Customer satisfaction acquisition methods based on facial expression recognition | |
CN108053398A (en) | A kind of melanoma automatic testing method of semi-supervised feature learning |
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 |
Application publication date: 20180424 |
|
RJ01 | Rejection of invention patent application after publication |