The content of the invention
To realize above-mentioned technical purpose, above-mentioned technique effect is reached, the present invention is achieved through the following technical solutions:A kind of base
In the timber visual identity method of multiple perceptron, it is characterised in that comprise the following steps:
Step 1:Founding mathematical models simultaneously carry out sample training, so as to generate multilayer perceptron parameter;
Step 2:Target enters cog region, and wood surface is taken pictures, and extracts clarification of objective value;
Step 3:Characteristic value is input into multilayer perceptron, is identified, if target can not be recognized, carry out abnormality processing, protected
Found exceptional sample is stayed, new samples are added in database, new training sample set of laying equal stress on, it is ensured that the similar spy that next time occurs
Different situation can be accurately identified;If target can be recognized, result to executing agency is sent;
Step 4:Executing agency carries out sort operation to target according to the result for receiving, and return to step 2 is ready to carry out down
The sort instructions of one target.
Wherein, the Mathematical Modeling of the multilayer perceptron includes input layer, hidden layer and output layer, wherein, the input
Layer includes system input, predominantly object feature value;Hidden layer includes weighted sum, excitation function.
The clarification of objective value is extracted and is captured and calculated by vision system, carries out image procossing to timber first,
Then many texture informations based on gray scale are extracted by calculating the gray level co-occurrence matrixes of timber.
The clarification of objective value is extracted and is captured and calculated by vision system, carries out image procossing to timber first,
Then many texture informations based on gray scale are extracted by calculating the gray level co-occurrence matrixes of timber.
The situation of prior art is different from, the beneficial effects of the invention are as follows:By automated system, information system and regard
The fusion of feel system, the system eliminates the process of manual sort's timber, improves the accuracy of production efficiency and classification, while
Or quality is related there is provided guarantee.
The method cost of implementation is low, and with very strong flexibility and scalability, the accuracy of whole system can be in operation
During improve constantly.The enterprise related for timber processing, the method can for its a set of feature based data are provided can
The discrimination standard of quantization.Enterprise can be on this system realized from growth according to the actual conditions of oneself, in reduction human cost
The economic worth that producing line is brought is improved simultaneously.
Specific embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
Referring to Fig. 1, there is provided a kind of timber visual identity method based on multiple perceptron, it is characterised in that including as follows
Step:
Step 1:Founding mathematical models simultaneously carry out sample training, so as to generate multilayer perceptron parameter;
Step 2:Target enters cog region, and wood surface is taken pictures, and extracts clarification of objective value;
Step 3:Characteristic value is input into multilayer perceptron, is identified, if target can not be recognized, carry out abnormality processing, protected
Found exceptional sample is stayed, new samples are added in database, new training sample set of laying equal stress on, it is ensured that the similar spy that next time occurs
Different situation can be accurately identified;If target can be recognized, result to executing agency is sent;
Step 4:Executing agency carries out sort operation to target according to the result for receiving, and return to step 2 is ready to carry out down
The sort instructions of one target.
Specifically,
The Mathematical Modeling of the multilayer perceptron includes input layer, hidden layer and output layer, specifically, input layer is mainly wrapped
System input is included, herein i.e. characteristic vector.Hidden layer includes the contents such as weighted sum, excitation function, the output of last layer
Used as next layer of input, can finally adjust out one group of linear coefficient w and b, and this tuning process can be traditional BP algorithm
Or other optimized algorithms.Linear result of calculation is produced output by the processing unit of hidden layer by nonlinear activation function.Finally
Result would generally use Function Mapping to 0, between 1, that is, provide the probability of every kind of classification.The generation of hidden layer and processing unit is
The process being trained to the input sample of known output, training result is one group of parameter, is usually implemented as file and is stored in meter
In calculation machine system.Training sample is more, then system classifies more accurate to clarification of objective.Factory can be to existing before online implementing
Product is trained, and produces corresponding multilayer perceptron training file.The clarification of objective value is extracted by vision system capture simultaneously
Calculate, image procossing is carried out to timber first, then extract many by calculating the gray level co-occurrence matrixes of timber
Texture information based on gray scale, what is extracted states texture information including target energy, correlation, unfavourable balance away from, contrast and image
Clarification of objective value described in entropy is extracted and is captured and calculated by vision system, can be envisioned as an one-dimensional vector of 1*M
FV=[X0,X1,X2….Xm], these features can be timber entirety or local feature.Important for one, timber is characterized in it
Texture.Its texture feature vector is obtained, first has to carry out image procossing to timber.By the gray scale symbiosis square for calculating timber
Battle array, we can therefrom extract many useful informations on texture.
Gray level co-occurrence matrixes are two pixel gray scales i, j on θ directions at a distance of d, i.e. (x, y) and (x+a, x+b) in image
The matrix that probability element p (i, j) of appearance is constituted.By gray matrix, we can extract many textures based on gray scale
Information, resulting texture information include energy, correlation, unfavourable balance away from, contrast, image entropy, while having an anisotropy that, ash
Degree histogram, colouring information etc., these characteristic values carry out spy as the mark to material by the K sample to the timber that has no guts
Extraction is levied, we most obtain N × K characteristic vector at last, altogether N × K × M feature, for same timber, their spy
Levying vector has similitude, and for different timber, their characteristic vector relevance will be reduced, meanwhile, in order to avoid occurring
The problem of over-fitting and dimension disaster, feature M should not take excessively.
Through the above way, whole system is in the process of running have very strong autgmentability and self-optimized ability,
Constantly during operation, the identification accuracy of system can be improved constantly, even if due to certain special case of certain type timber
There is accidental None- identified or identification mistake, being run into equally be can guarantee that after wood feature addition Sample Storehouse so in next time
Can be accurately identified during special circumstances.And due to just having carried out enough sample trainings before system startup, such situation occurs
Probability is inherently very low, and system can be with stable operation in the starting stage.
Embodiments of the invention are the foregoing is only, the scope of the claims of the invention is not thereby limited, it is every to utilize this hair
Equivalent structure or equivalent flow conversion that bright specification and accompanying drawing content are made, or direct or brief introduction are used in other correlation techniques
Field, is included within the scope of the present invention.