CN104331704A

CN104331704A - Plant identification method based on Haar characteristics

Info

Publication number: CN104331704A
Application number: CN201410580980.7A
Authority: CN
Inventors: 胡平
Original assignee: HEFEI XINGFU INFORMATION TECHNOLOGY Co Ltd
Current assignee: HEFEI XINGFU INFORMATION TECHNOLOGY Co Ltd
Priority date: 2014-10-27
Filing date: 2014-10-27
Publication date: 2015-02-04

Abstract

A plant identification method based on Haar characteristics includes the steps: 1 establishing a plant type Haar characteristic knowledge base; 2 inputting pictures and text messages of tested samples, extracting characteristics of the tested samples, and respectively computing Haar characteristic value arrays of characteristic vectors of the tested samples on fifteen dimensions. For a certain characteristic vector, if the overlapping ratio of the samples and model values on the fifteen dimensions is smaller than a preset threshold value, matching of the characteristics is failed, and a plant type matched in a model base is a search result when all characteristics are successfully matched.

Description

A kind of plants identification method based on Haar feature

Technical field

The invention belongs to CRT technology field, particularly relating to a kind of method of carrying out plants identification by extracting plant leaf Haar feature.

Background technology

Along with the fast development of computer pattern recognition, by intelligent means assist people carry out plant identification and analyze become increasing Man's Demands.Except the professional of botany aspect, we mainly still rest on the experience stage to the identification of plant at present, are often in the ocean of a slice plant, are not but the name of several plant, let alone relevant knowledge, only have the plant be once familiar with just to can say simple characteristic.Also have some cell phone softwares that us can be assisted to carry out the identification of plant recently, the blade picture mainly through plant is analyzed.But just for the global shape of blade, also consideration petiole useless and stipule situation simultaneously, cause identification error comparatively large, the effect of identification is very undesirable.

Summary of the invention

Carry out a method for plants identification according to leaf feature, described method is, 1, set up floristics Haar feature knowledge storehouse; 2, input sample picture and Word message, feature extraction is carried out to sample, calculate the Haar eigenwert array of testing sample proper vector in 15 dimensions respectively; 3, for certain proper vector, if sample is less than default threshold value with the Duplication of model value in 15 dimensions, this characteristic matching failure.

Set up floristics Haar feature knowledge storehouse, be exactly, for every Plants sets up the other model of cognition of Haar feature, as Fig. 1, each feature is 15 dimensional feature vector a(haar_x2, haar_y2, haar_x3, haar_y3, haar_x2_y2,

haar_x4,haar_y4,haar_point,tilted_haar_x2,tilted_haar_y2,tilted_haar_x3,tilted_haar_y3,

Tilted_haar_x4, tilted_haar_y4, tilted_haar_point), the corresponding one group of Haar eigenwert of the every one dimension of vector.First the plant leaf having obvious feature is chosen, carry out Haar feature extraction, the namely integral operation of gray level image, calculate the eigenwert array of each dimension of each proper vector and preserve, morphogenesis characters sorter, set sample Duplication threshold value, the sample being less than threshold value is considered to not meet feature simultaneously.

Input sample picture and Word message, carry out feature extraction to sample, comprise manual input information and automatic discriminance analysis.Manual input information comprises petiole CF, the phyllotaxy of stipule color and quantity and blade and compound leaf number etc.The pro and con of sample blade to be analyzed is taken respectively the picture input system of generation, extract the eigenwert array of all dimensions of its Haar proper vector by system automatic analysis.

For certain proper vector (as end of blade proper vector), calculate the eigenwert array of testing sample in 15 dimensions, calculate the Duplication of eigenwert array in these eigenwert arrays and model of cognition, think equal within eigenwert difference 5%, Duplication height proves that characteristic matching is obvious, Duplication low proof characteristic matching is not obvious, lower than the threshold value of setting, thinks that characteristic matching is failed.The average Duplication of characteristic matching of 15 components of certain proper vector, more than setting threshold value (as 80%), thinks that this characteristic matching is successful.When all features, all the match is successful, the vegetation type mating out in model bank is exactly the result of our retrieval, if there is Partial Feature Matching failure, but most of characteristic matching success, the threshold value (as 80%) totally exceeding setting also thinks that the match is successful, if most of feature does not all have, the match is successful, thinks that current identifying operation is failed.

Innovative point of the present invention is:

1, adopt the Duplication of Haar eigenwert array to calculate character symbol right, greatly reduce the complexity of calculating;

2, blade, petiole, stipule three aspects are divided to identify on leaf;

3, blade is divided into again the features such as end of blade, leaf margin, phyllopodium, vein, phyllotaxy and compound leaf to identify;

4, artificial input and procedure identification combine, and maximize favourable factors and minimize unfavourable ones, and have complementary advantages.

Accompanying drawing explanation

Fig. 1 leaf Haar characteristic model;

Fig. 2 leaf identification process figure.

Embodiment

Describe the present invention in detail below in conjunction with accompanying drawing, it illustrates principle of the present invention as the part of this instructions by embodiment.

We choose a complete leaf sample, and sample class is unknown, by following steps identification:

First typing essential information is as follows:

1, input: petiole color: green; Shape: carefully cylindrical

2, input: stipule color: green; Quantity: 2

3, input: blade phyllotaxy: to life; Compound leaf: 2

Secondly, we gather a slice blade, take the picture with the back side above it.

Next, this two pictures is analyzed:

1, the color on automatic analysis leaf two sides.

2, end of blade Haar proper vector is analyzed.Suppose that its proper vector is A(A1, A2 ..., A15), A1, A2 ..., A15 is 15 features of Haar feature, and each feature is an eigenwert array.Here A1={-199 is supposed ,-127 ,-30,20,80,100 ....

The short sharp proper vector of 3, taking out the default end of blade of certain kind (as pea) from feature model library is B(B1, B2 ... B15), its characteristic component B1={-199,-127,-30,10,80,101 ..., if A1's and B1 has the Duplication exceeding threshold value, just think that sample has the feature of the B1 component of model (as pea), when sample characteristic vector A in institute important can with B in correspondence component have good Duplication, just think that sample has the B feature of model (as pea), i.e. short sharp feature; The subset S1 of the model composition model bank of all coupling sample characteristics, is shown in Fig. 2.

4, in like manner form S2 in S1 collective analysis phyllopodium feature, then form S3 at S2 collective analysis leaf margin characteristics, on S3 subset basis, finally analyze vein feature form S4.When all features all meet or most of feature all meets, just think that the kind of this sample is exactly the kind (as pea) of model.If S4 has multiple model, system all lists these models identified for artificial last selection.If do not matched, namely S4 is empty, just thinks current recognition failures.

Above disclosedly be only the preferred embodiments of the present invention, certainly can not limit the interest field of the present invention with this, therefore according to the equivalent variations that the present patent application the scope of the claims is done, still belong to the scope that the present invention is contained.

Claims

1. carry out a method for plants identification according to leaf feature, wherein, described method is:

1) floristics Haar feature knowledge storehouse, is set up.

2), input sample picture and Word message, feature extraction is carried out to sample, calculate the Haar eigenwert array of testing sample proper vector in 15 dimensions respectively.

3), for certain proper vector, if sample is less than default threshold value with the Duplication of model value in 15 dimensions, this characteristic matching failure.

4), when all features, all the match is successful, the vegetation type mating out in model bank is exactly the result of our retrieval, if there is Partial Feature Matching failure, but most of characteristic matching success, the threshold value totally exceeding setting also thinks that the match is successful, if most of feature does not all have, the match is successful, thinks that current identifying operation is failed.