CN102760228A - Specimen-based automatic lepidoptera insect species identification method - Google Patents
Specimen-based automatic lepidoptera insect species identification method Download PDFInfo
- Publication number
- CN102760228A CN102760228A CN2011101069551A CN201110106955A CN102760228A CN 102760228 A CN102760228 A CN 102760228A CN 2011101069551 A CN2011101069551 A CN 2011101069551A CN 201110106955 A CN201110106955 A CN 201110106955A CN 102760228 A CN102760228 A CN 102760228A
- Authority
- CN
- China
- Prior art keywords
- image
- wing
- insect
- background
- right sides
- 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.)
- Granted
Links
Images
Abstract
The invention relates to a specimen-based automatic lepidoptera insect species identification method. The method comprises the following steps: removing background by using LazySnapping and/or GrabCut, segmenting left and right wings by using a projection-based method, and determining critical points by using an edge fitting method and aligning the positions of the left and right wings; carrying out feature extraction in three RGB (Red, Green, Blue) channels, dividing a wing part into a plurality of cells in radius and angle directions, and calculating pixel average values of each cell to construct final feature vectors; and realizing the classification identification by using an SVM (Support Vector Machine) classifier. The lepidoptera insect specimen image identification method is simple and convenient to operate, high in recognition precision and strong in fault tolerance, has ideal time performance, and can be used for obviously improving the lepidoptera insect species identification efficiency.
Description
Technical field
The present invention relates to a kind of caste automatic identification method based on the number of samples word image; Particularly to the automatic evaluation of lepidopterous insects; Can be applicable to fields such as plant quarantine, plant pest prediction and control thereof, or can be used as reference and the reference that important component part is used for ecological information science research.This technology can be adopted by departments such as customs, plant quarantine department, the agricultural prevention and control of plant diseases, pest control.Can be the means that the grass-roots work personnel that do not possess relevant professional or peasant provide automatic discriminating.
Background technology
Lepidoptera (Lepidoptera) is a second largest order in the Insecta, gains the name owing to having draped over one's shoulders a large amount of scales on health and the wing.Mostly lepidopterous larvae is phytophagous, and the harm that crops are caused is bigger, the lepidoptera imago host of generally not causing harm, but the part insect is arranged as inhaling fruit leaf moth, the beak point punctures pericarp and draws juice, and fruit is worked the mischief.Therefore, identify that effectively Lepidoptera is significant to the control of disease and pest.
Traditional classification of insect only is confined to minority insect researcher, plant protection scientific worker with identification, relies on manual inspection, visual inspection morphological feature to carry out, and needs the accumulation of solid classification of insect knowledge and experience; Except that the minority expert; Common people are difficult to grasp, and the subjective mood that the result is generally the discriminator influences the discrimination instability; Especially under the situation long in the time, that workload is big, False Rate increases.Detection can not Real-time and Dynamic carry out, need a large amount of artificial, sampled point is not enough, this to a great extent limit the popularization degrees of people to the understanding of insect, bring tremendous loss usually for agricultural production and economic activity.Computer vision such as advanced person's Flame Image Process, pattern-recognition and signal processing technology are applied to classification of insect and identification field, realize the automatic identification of insect, for the popularization degree that promotes the human knowledge; Reduce in the agricultural production because the disease that insect causes, avoid economic loss; Effectively the rare species of insect of protection is safeguarded the ecologic environment balance, has the effect of can not ignore.
Digital image processing techniques are Along with computer technology and generation, the development of VLSI (VLSI (very large scale integrated circuits)) and an emerging technology field of constantly growing up sixties in 20th century, and obtain bigger development in the nineties.Be used widely in fields such as Aero-Space, living things feature recognition, biomedical engineering, machine vision, multimedia messages processing, and obtain good effect.But digital image processing techniques are applied to the research in insect identification field just just has been in the starting stage, and relevant document is still rare, on technology realizes, is still waiting further exploitation and perfect.
The research that the insect mathematical morphology combines with computer technology is mainly since the nineties in 20th century.British government initiated DAISY (Digital Automated Identification SYstem) research engineering in 1996, in worldwide, had started the upsurge of the automatic Study of recognition of relevant insect.Research and exploration through more than ten years; At present the most representative insect automatic recognition software have exploitation such as Steinhage exploitations such as ABIS (Automatic Bee IdentificationSystem), Weeks DAISY and and the SPIDA (SPeciesIDentitfied Automatically) of Russell and Martin exploitation, the software DrawWing of exploitation such as Tofilski etc.
At home; The IPMIST that professor Shen Zuorui of China Agricultural University instructs (the ecological intellectual technology of plant protection system) laboratory utilized the means of computer patterns identification to carry out caste and identifies since 1997; And launching more deep discussion aspect the mathematical morphology of IMAQ, utilization and the insect image of insect; Utilize neural network that insect has been carried out automatic discriminating, automatically measure and kind has obtained a series of progress aspect the evaluation automatically in the insect form.Zhao's historical records etc. are realized differentiating automatically to 40 kinds of insects with 11 mathematics morphological features such as polypide area, girth; Yao Qing etc. are index with finned surface orthogonal projection periphery curvature, and five kinds of moth classes are migrated, and insect reaches and the fore wing wing shape of the non-insect that migrates that it is approximate is carried out numerical analysis and comparison; Huang Shiguo etc. have also carried out systematic research to the insect identification gordian technique based on image; Insect recognizer to based on shape facility, textural characteristics and morphology, body is studied; But its experiment data set only comprises five types of insect samples, and the result is still waiting to prove with experimental data more fully.
Above-mentioned result of study has plenty of based on the vein characteristic; Have plenty of based on shape facility; The vein characteristic is more suitable to hymenopteran, and lepidopterous insects will extract the vein characteristic, at first will use chemical method to remove the scale and the color spot of finned surface; Get access to butterfly vein picture through scanning, process is complicated and damage sample easily; Shape facility is then responsive to the attitude of insect, dimensional variation, and if insect specimen just be difficult to correct identification when incompleteness is arranged.The result of study of majority method all is based on the small sample storehouse test of limited type of insect, and at recognition correct rate, aspects such as recognition time performance are still waiting further checking, also do not reach the degree that can directly apply to production.
Summary of the invention
The object of the present invention is to provide a kind of method of automatic identification lepidopterous insects image.It mainly solves by the insect image pattern and realizes lepidopterous insects kind automatic recognition problem through computer pattern recognition.Particularly excalation is arranged at the insect sample, or the sample attitude in addition the distance in limited range, change, still can effectively identify the caste of notable feature with superior performance.Insect specimen need not use chemical method to remove the scale and the color spot of finned surface, avoids the existing complex process of bringing based on the method for vein characteristic.And the insect recognition methods that solves based on the picture shape characteristic changes the precision property decline that produces to fragmentary sample, graphical rule.
The technical scheme that the present invention adopts is:
A kind of lepidopterous insects kind automatic identification method based on the sample image may further comprise the steps:
1) image pre-service
Remove the background of Lepidoptera sample coloured image, the image after the removal background is carried out carrying out binaryzation behind gray processing, the gaussian filtering, carry out left and right sides wing again and cut apart;
Aim at wing position, the left and right sides, cut the square area that comprises effective wing zone;
Lepidoptera sample coloured image is done same conversion, that is, cut apart left and right sides wing earlier, aim at wing position, the left and right sides again, cut the square area that comprises effective wing zone, obtain rotating the left wing coloured image and the right wing coloured image of alignment at last;
2) image characteristics extraction
Left and right sides wing coloured image after the position alignment is decomposed to three passages of RGB; Identical characteristic extraction procedure below each passage carried out: the finned surface image is divided between several region along radius and angle direction, calculates each finned surface pixel average in interval, the pixel average series connection that wing three all intervals of passage in the left and right sides are calculated; Obtain the proper vector of insect general image; The proper vector element value is carried out convergent-divergent, normalize to [0,1] interval;
3) classification is differentiated
Confirm training set earlier, the retraining sorter model carries out Classification and Identification at last.
Further:
In said step 1), use one of following method to remove the background of sample image:
Remove the background of sample image with Lazy snapping method; Method is in the foreground area that needs keep, to mark with a kind of lines of color; In the background area that needs are removed, mark with the lines of another kind of color; Lazy Snapping algorithm calculates the separatrix between prospect and the background automatically, if cut apart not enough accurately then the fine setting of marking repeatedly meets the requirements until the separatrix;
Or remove the background of sample image with the Grabcut instrument, method is that the minimum rectangle frame that comprises foreground area is set, and black is arranged in the background area after cutting apart completion;
Or with the work of GrabCut+Lazy Snapping instrument completion background removal; Method is to sketch the contours of foreground area with GrabCut earlier; And then the prospect removed of the background of not removing with Lazy Snapping mark and mistake, black is arranged in the background area after cutting apart completion.
In said step 1), the image after the removal background is carried out gray processing be meant that the employing method of weighted mean obtains gray level image.
RGB three-component in the image after the removal background carries out weighted mean by Y=0.299*R+0.587*G+0.114*B.
In said step 1), said binaryzation is promptly selected a gray threshold, is set to white greater than the pixel of threshold value, is set to black less than the pixel of threshold value.
In said step 1), said left and right sides wing dividing method is that the bianry image that forms is got largest contours; This contour images is therefrom asked projection in the mind-set both sides; Obtaining two local minizing points, is the boundary with the horizontal ordinate of these two local minizing points, is partitioned into the wing zone, the left and right sides of insect.
In said step 1), the method for said left and right sides wing position alignment is, search coboundary and lower limb are used fitting a straight line to last lower limb on the profile diagram of left and right sides wing, and laying one's heart bear with two straight lines is the center image rotating, until the upper edge level.
In said step 3); The method of said definite training set is that the corresponding support vector machine classifier of each type insect is with the positive example of several samples conducts of this type insect; Several samples of other types insect are as negative example, set by step 2) method is extracted the proper vector of each type insect.
In said step 3), the method for said training classifier model is to use positive example and counter-example proper vector training support vector machine classifier model that a step normalization obtains, the corresponding sorter model of each type insect.
In said step 3); The method of said Classification and Identification is, with the insect specimen image of unknown classification set by step 1) and 2) carry out pre-service and feature extraction after, with the input of proper vector as each support vector machine classifier; If the output valve of certain class support vector machines sorter is for just; Then be accepted as this type insect, if output valve then is judged as non-class insect for negative.The invention has the advantages that: lepidopterous insects automatic distinguishing method for image of the present invention, need not remove surperficial scale and color spot with chemical reagent to the insect image, image-pickup method is simple and easy to operate.In preprocessing process, can remove variations such as translation, rotation, the proper vector of constructing during feature extraction not only has the yardstick unchangeability, and fault-tolerance is preferably also arranged when sample had excalation.Single sorter is carried out the classification time below 50ms, and the accuracy of classification is more than 90%.
Description of drawings
Fig. 1 is the former figure of sample image;
Fig. 2 is the sample image after the removal background from Fig. 1;
Fig. 3 is a profile diagram;
Fig. 4 is for cutting apart left and right sides wing area schematic;
Fig. 5 is with fitting a straight line coboundary and lower limb synoptic diagram;
Fig. 6 is for rotating the left and right sides wing profile diagram after aliging;
Fig. 7 is for rotating the left and right sides wing image after aliging;
Fig. 8 is cut apart signal for the interval;
Fig. 9 is for removing background with Lazy Snapping instrument;
Figure 10 is for removing background with the Grabcut instrument.
Embodiment
The present invention includes following steps:
1) image pre-service: remove the background of Lepidoptera sample coloured image, the image after the removal background is carried out carrying out binaryzation behind gray processing, the gaussian filtering, carry out left and right sides wing again and cut apart; Aim at wing position, the left and right sides, cut the square area that comprises effective wing zone; Lepidoptera sample coloured image is done same conversion, that is, cut apart left and right sides wing earlier, aim at wing position, the left and right sides again, cut the square area that comprises effective wing zone, obtain rotating the left wing coloured image and the right wing coloured image of alignment at last.
2) image characteristics extraction: the left and right sides wing coloured image after the position alignment is decomposed to three passages of RGB; Identical characteristic extraction procedure below each passage carried out: the finned surface image is divided between several region along radius and angle direction, calculates each finned surface pixel average in interval, the pixel average series connection that wing three all intervals of passage in the left and right sides are calculated; Obtain the proper vector of insect general image; The proper vector element value is carried out convergent-divergent, normalize to [0,1] interval.
3) classification is differentiated: confirm training set earlier, the retraining sorter model carries out Classification and Identification at last.
Specify below in conjunction with accompanying drawing.
1) image pre-service
Use digital camera to take the Lepidoptera sample; Obtain the original color image of Lepidoptera sample, remove background, background color is set (is shown as white among Fig. 2 for monochromatic with Lazy snapping method; Reality background usually is set to black), prospect keeps former figure information.Image after original image and the removal background is seen Fig. 1, Fig. 2.
Image to removing after the background carries out gray processing, gaussian filtering, carries out binaryzation again, and the result of binaryzation is got largest contours.Wherein:
So-called largest contours is meant one that in bianry image, takes out the area maximum in the detected profile, can obtain insect contour images as shown in Figure 3.
Gray processing is about to the RGB coloured image and converts monochrome image into, and the present invention adopts method of weighted mean, because human eye is the highest to the sensitivity of green, to blue responsive minimum, therefore, by following formula the RGB three-component carried out weighted mean and can obtain reasonably gray level image:
Y＝0.299*R+0.587*G+0.114*B
The purpose of gaussian filtering is a smoothing denoising, and the binaryzation operating result is opened prospect and background separation as far as possible.
Binaryzation is promptly selected a gray threshold; Pixel greater than threshold value is set to white (being that the pixel value size is 255); Pixel less than threshold value is set to black (being that the pixel value size is 0); Because of being set to black in this step of background removal background, threshold value is set to a smaller value 20 here.In addition,, there is the situation that the subregion two-value of insect finned surface is turned to black, therefore eliminates the isolated area in the finned surface zone with the method for extracting largest contours because the gray scale texture variations of insect finned surface is more.
Therefrom ask vertical projection in the mind-set both sides for this contour images; Can obtain two local minizing points; Vertical projection is promptly added up the number of foreground pixel on each X coordinate figure along Y direction; Minimum point i.e. this X coordinate figure place foreground pixel aggregate-value is minimizing X coordinate, like two black vertical line positions on the trunk both sides among Fig. 4.Horizontal ordinate with these two local minizing points is the boundary, can be partitioned into the wing zone, the left and right sides of insect roughly, like Fig. 4.Follow the tracks of contour edge from the partitioning boundary of left wing and right wing; Coboundary begins to follow the tracks of the point until the peak b of profile from the last summit a point on border; Lower limb begins to follow the tracks of the point until the minimum point d of profile from the following summit c point on border; With least square method the coboundary that is in left wing and right wing and the pixel on the lower limb are used fitting a straight line, and ask for the intersection point O of the edge line up and down of left wing and right wing respectively, like Fig. 5.Then left wing being rotated to the upper edge straight line around two straight-line intersection O is level, and right wing is also made same treatment.Cut the square area that comprises effective wing zone again, like Fig. 6.
The central point that can confirm to rotate through black white image and the angle of rotation, and the effective coverage of finned surface, thus cut out corresponding color image region, like this coloured image only needs can cut apart left wing and the right wing that obtains rotating alignment, like Fig. 7.
2) image characteristics extraction
Feature Extraction is carried out respectively on three passages of RGB.At first coloured image shown in Figure 7 is transformed to three passages of RGB, obtain three width of cloth gray level images.On this three width of cloth gray level image, extract characteristic respectively, the finned surface image is divided into M and N zone, M=5 among the present invention at radial direction and angle direction respectively; N=8; So whole finned surface can be divided into 40 intervals (like Fig. 8), calculates each interval pixel average, is about to interval interior pixel value and adds up the back divided by the number of pixels in interval; Can to obtain length be 40 vector to each passage so; Obtain the vector of 40*3=120 after the vector series connection of three passages, each 120 dimension of left wing and right wing can obtain the proper vector that 120*2=240 ties up from the insect general image so.
3) classification is differentiated
Confirm training set earlier, the retraining sorter model carries out Classification and Identification at last.
The method of confirming training set is, the corresponding support vector machine classifier of each type insect, as positive example, several samples of other types insect are as negative example with several samples of this type insect, set by step 2) method extract the proper vector of each type insect.
The method of training classifier model is to use positive example and counter-example proper vector training support vector machine classifier model that a step normalization obtains, the corresponding sorter model of each type insect.
The method of Classification and Identification is; With the insect specimen image of unknown classification set by step 1) and 2) carry out pre-service and feature extraction after; With the input of proper vector, if the output valve of certain class support vector machines sorter then is accepted as this type insect for just as each support vector machine classifier; If output valve then is judged as non-class insect for negative.
Classification and the work of differentiating are accomplished by sorter.A last step is extracted the proper vector obtain to be normalized to the interval { 0,1} is promptly to vectorial X={x
1, x
2..., x
nCarry out as follows and operate:
x′
i＝x
i/255 i＝1，2，...，n
x
i(i=1,2 ..., n), x
iI the element that refers to feature vector, X.Because each pixel span of gray level image is 0～255; Span as each interval pixel average also is 0～255; Through the calculating of following formula, can normalize to scope 0～1 to the value of each element in the proper vector, thereby obtain normalized feature vector, X '=x '
1, x '
2..., x '
n.
The corresponding sorter of each type insect, (Support Vector Machine SVM) realizes classification by SVMs in the present invention.For each svm classifier device, followed identical principle of design: at first adopting has the method for supervised training to train, and selects suitable kernel function and parameter; Use test test sample discrimination then.The selection of kernel function and parameters of choice are numerous and diverse processes, but Taiwan
Lin ZhirenDoctor uses
PythonThe shell script of exploitation can be realized Automatic parameter selection to obtain optimum cross validation precision, uses this grid.py script requirement installation
PythonWith
Gnuplot..
The characteristic that this type insect is extracted in the sample set is routine as just with gathering; Positive example is meant that sorter is accepted as the sample of this type (mark be output as+1); In other classification insects, extract characteristic as negative example (mark is output as-1), train the corresponding sorter model of every type of insect.
Can type of carrying out build-in test with the model that obtains of training to the test sample book of similar insect, when being output as positive number, think acceptance, be output as just non-ly, think refusal.
Model with training obtains can be tested between type of carrying out any test sample book of inhomogeneity insect, when being output as positive number, accepts, and is just non-, refusal.
Test set of the present invention comprises 10 types of lepidopterous insects adult general images, and every type of insect collection sample image 48～66 width of cloth wherein 4/5 are used for training, and 1/5 is used for testing.The svm classifier device of training polynomial kernel function; The gamma value of this kernel function is 1, and the coef0 in the kernel function is set to 1, and the degree in the kernel function is set to 3; The model that uses Svmtrain to train is tested; The accuracy of coupling is average 91.96% in the class as a result, and the coupling accuracy is 91.77% between type, and average recognition rate is 91.79%.
The instance that combines concrete implementation method below again, further detailed description is done in the automatic discriminating of lepidopterous insects sample image of the present invention:
Instance 1
1. use " nEO iMAGING " subsidiary stingy figure functional module, accomplish the background removal work from Fig. 1 to Fig. 2, and be arranged to black to background.
2. open the Lepidoptera sample general image file of removing after the background, the image of opening to be identified is presented at the window upper left side.
3. click [coupling] Submenu Items under [operation] menu, will start whole evaluation matching process, qualification result is presented at the window right side, presses the precedence output qualification result information that sorter is accepted, insect title and the sample image promptly identified.Generally speaking, because the corresponding sorter model of every type of insect in the storehouse has only one, so correct evaluation output result has only one; When existing erroneous judgement as a result, piece image possibly accepted by a plurality of sorters simultaneously, exports the result this moment by first three received classification of matching order output.
4. if training classifier model again; Can click [training] Submenu Items under [operation] menu; In the window that ejects; Can select parameters such as path, training data file place, pattern number, svm classifier device type, kernel function type and cost, gamma value, can train again after point is confirmed and generate the model of specifying numbering.
Instance 2
Accomplish background removal work with Lazy Snapping instrument from Fig. 9 a to Fig. 9 b; Method is in the foreground area that needs keep, to mark with a kind of lines of color; In the background area that needs are removed, mark with the lines of another kind of color; Lazy Snapping algorithm will calculate the separatrix between prospect and the background automatically, if cut apart not enough accurately then the fine setting of marking repeatedly meets the requirements until the separatrix.Lazy Snapping is a kind of focus object partitioning algorithm, and during operation, the user drags mouse and comes some lines of mark through pinning button (left button display foreground, right button display background).Yellow line mark prospect for example, blue line mark background.This high-level drawing formula user interface does not need point-device user's input.Trigger cutting procedure through clicking preview button after having drawn markings, then, the segmentation result on the customer inspection screen, and whether decision needs the more line of mark.The partitioning boundary error that this method produces is very little, through increasing the color similarity degree on object inside and border to greatest extent, comes the optimization objects border.Square frame among Fig. 9 a is the little scraps of paper in the background when entomologizing image, and the instrument of Lazy Snapping algorithm can be effectively this background removal.
Follow-up evaluation operation is with the 2.3rd step of instance 1. and training process is with the 4th step of instance 1.
Instance 3
Accomplish background removal work with the Grabcut instrument from Figure 10 a to Figure 10 c; Square frame among Figure 10 a is the little scraps of paper in the background when entomologizing image; The rectangle frame of Figure 10 b is the artificial minimum rectangle frame that comprises foreground area that is provided with, and black is arranged in the background area after cutting apart completion.
Follow-up evaluation operation is with the 2.3rd step of instance 1. and training process is with the 4th step of instance 1.
Instance 4
Accomplish background removal work with GrabCut+Lazy Snapping instrument.For the special complex image of background, can sketch the contours of foreground area roughly with GrabCut earlier, and then the prospect removed of the background of not removing with Lazy Snapping mark and mistake, black is arranged in the background area after cutting apart completion.
Follow-up evaluation operation is with the 2.3rd step of instance 1. and training process is with the 4th step of instance 1.
Adopt above four kinds of instances in comprising 10 types of lepidopterous insects sample image pattern storehouses, to carry out discrimination test, all obtained good test effect.
Claims (10)
1. lepidopterous insects kind automatic identification method based on the sample image is characterized in that may further comprise the steps:
1) image pre-service
Remove the background of Lepidoptera sample coloured image, the image after the removal background is carried out carrying out binaryzation behind gray processing, the gaussian filtering, carry out left and right sides wing again and cut apart;
Aim at wing position, the left and right sides, cut the square area that comprises effective wing zone;
Lepidoptera sample coloured image is done same conversion, that is, cut apart left and right sides wing earlier, aim at wing position, the left and right sides again, cut the square area that comprises effective wing zone, obtain rotating the left wing coloured image and the right wing coloured image of alignment at last;
2) image characteristics extraction
With the left and right sides wing coloured image after the position alignment decompose to three passages of RGB each passage is carried out below identical characteristic extraction procedure: the finned surface image is divided between several region along radius and angle direction; Calculate each interval interior finned surface pixel average; With the pixel average series connection that wing three all intervals of passage in the left and right sides calculate, obtain the proper vector of insect general image, the proper vector element value is carried out convergent-divergent; Normalize to [0,1] interval;
3) classification is differentiated
Confirm training set earlier, the retraining sorter model carries out Classification and Identification at last.
2. the lepidopterous insects kind automatic identification method based on the sample image according to claim 1 is characterized in that:
In said step 1), use one of following method to remove the background of sample image:
Remove the background of sample image with Lazy snapping method; Method is in the foreground area that needs keep, to mark with a kind of lines of color; In the background area that needs are removed, mark with the lines of another kind of color; Lazy Snapping algorithm calculates the separatrix between prospect and the background automatically, if cut apart not enough accurately then the fine setting of marking repeatedly meets the requirements until the separatrix;
Or remove the background of sample image with the Grabcut instrument, method is that the minimum rectangle frame that comprises foreground area is set, and black is arranged in the background area after cutting apart completion;
Or with the work of GrabCut+Lazy Snapping instrument completion background removal; Method is to sketch the contours of foreground area with GrabCut earlier; And then the prospect removed of the background of not removing with Lazy Snapping mark and mistake, black is arranged in the background area after cutting apart completion.
3. the lepidopterous insects kind automatic identification method based on the sample image according to claim 1 is characterized in that:
In said step 1), the image after the removal background is carried out gray processing be meant that the employing method of weighted mean obtains gray level image.
4. the lepidopterous insects kind automatic identification method based on the sample image according to claim 3 is characterized in that: the RGB three-component in the image after the removal background carries out weighted mean by Y=0.299*R+0.587*G+0.114*B.
5. the lepidopterous insects kind automatic identification method based on the sample image according to claim 1 is characterized in that:
In said step 1), said binaryzation is promptly selected a gray threshold, is set to white greater than the pixel of threshold value, is set to black less than the pixel of threshold value.
6. the lepidopterous insects kind automatic identification method based on the sample image according to claim 1 is characterized in that:
In said step 1), said left and right sides wing dividing method is that the bianry image that forms is got largest contours; This contour images is therefrom asked projection in the mind-set both sides; Obtaining two local minizing points, is the boundary with the horizontal ordinate of these two local minizing points, is partitioned into the wing zone, the left and right sides of insect.
7. the lepidopterous insects kind automatic identification method based on the sample image according to claim 1 is characterized in that:
In said step 1), the method for said left and right sides wing position alignment is, search coboundary and lower limb are used fitting a straight line to last lower limb on the profile diagram of left and right sides wing, and laying one's heart bear with two straight lines is the center image rotating, until the upper edge level.
8. the lepidopterous insects kind automatic identification method based on the sample image according to claim 1 is characterized in that:
In said step 3); The method of said definite training set is that the corresponding support vector machine classifier of each type insect is with the positive example of several samples conducts of this type insect; Several samples of other types insect are as negative example, set by step 2) method is extracted the proper vector of each type insect.
9. the lepidopterous insects kind automatic identification method based on the sample image according to claim 8 is characterized in that:
In said step 3), the method for said training classifier model is to use positive example and counter-example proper vector training support vector machine classifier model that a step normalization obtains, the corresponding sorter model of each type insect.
10. the lepidopterous insects kind automatic identification method based on the sample image according to claim 9 is characterized in that:
In said step 3); The method of said Classification and Identification is, with the insect specimen image of unknown classification set by step 1) and 2) carry out pre-service and feature extraction after, with the input of proper vector as each support vector machine classifier; If the output valve of certain class support vector machines sorter is for just; Then be accepted as this type insect, if output valve then is judged as non-class insect for negative.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110106955.1A CN102760228B (en) | 2011-04-27 | 2011-04-27 | Specimen-based automatic lepidoptera insect species identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110106955.1A CN102760228B (en) | 2011-04-27 | 2011-04-27 | Specimen-based automatic lepidoptera insect species identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102760228A true CN102760228A (en) | 2012-10-31 |
| CN102760228B CN102760228B (en) | 2014-04-30 |
Family
ID=47054683
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110106955.1A Expired - Fee Related CN102760228B (en) | 2011-04-27 | 2011-04-27 | Specimen-based automatic lepidoptera insect species identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102760228B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631451A (en) * | 2016-01-07 | 2016-06-01 | 同济大学 | Plant leave identification method based on android system |
| CN107292314A (en) * | 2016-03-30 | 2017-10-24 | 浙江工商大学 | A kind of lepidopterous insects species automatic identification method based on CNN |
| CN108109170A (en) * | 2017-12-18 | 2018-06-01 | 上海联影医疗科技有限公司 | Medical image scan method and medical imaging device |
| CN108734719A (en) * | 2017-04-14 | 2018-11-02 | 浙江工商大学 | Background automatic division method before a kind of lepidopterous insects image based on full convolutional neural networks |
| CN109271905A (en) * | 2018-09-03 | 2019-01-25 | 东南大学 | A kind of black smoke vehicle detection method based on single-frame images |
| CN109840549A (en) * | 2019-01-07 | 2019-06-04 | 武汉南博网络科技有限公司 | A kind of pest and disease damage recognition methods and device |
| EP3494786A1 (en) | 2017-12-07 | 2019-06-12 | Bayer CropScience Aktiengesellschaft | Detection of pests |
| CN110689039A (en) * | 2019-08-19 | 2020-01-14 | 浙江工业大学 | Trunk texture identification method based on four-channel convolutional neural network |
| US11877873B2 (en) | 2017-12-18 | 2024-01-23 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for determining scanning parameter in imaging |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726251A (en) * | 2009-11-13 | 2010-06-09 | 江苏大学 | Automatic fruit identification method of apple picking robot on basis of support vector machine |
-
2011
- 2011-04-27 CN CN201110106955.1A patent/CN102760228B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726251A (en) * | 2009-11-13 | 2010-06-09 | 江苏大学 | Automatic fruit identification method of apple picking robot on basis of support vector machine |
Non-Patent Citations (2)
| Title |
|---|
| 唐强: "基于图像模式识别技术的昆虫识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 王媛媛: "图像分割和边缘检测方法在昆虫图像中的应用", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631451A (en) * | 2016-01-07 | 2016-06-01 | 同济大学 | Plant leave identification method based on android system |
| CN107292314A (en) * | 2016-03-30 | 2017-10-24 | 浙江工商大学 | A kind of lepidopterous insects species automatic identification method based on CNN |
| CN108734719A (en) * | 2017-04-14 | 2018-11-02 | 浙江工商大学 | Background automatic division method before a kind of lepidopterous insects image based on full convolutional neural networks |
| EP3494786A1 (en) | 2017-12-07 | 2019-06-12 | Bayer CropScience Aktiengesellschaft | Detection of pests |
| CN108109170A (en) * | 2017-12-18 | 2018-06-01 | 上海联影医疗科技有限公司 | Medical image scan method and medical imaging device |
| CN108109170B (en) * | 2017-12-18 | 2022-11-08 | 上海联影医疗科技股份有限公司 | Medical image scanning method and medical imaging equipment |
| US11877873B2 (en) | 2017-12-18 | 2024-01-23 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for determining scanning parameter in imaging |
| CN109271905A (en) * | 2018-09-03 | 2019-01-25 | 东南大学 | A kind of black smoke vehicle detection method based on single-frame images |
| CN109271905B (en) * | 2018-09-03 | 2021-11-19 | 东南大学 | Black smoke vehicle detection method based on single-frame image |
| CN109840549A (en) * | 2019-01-07 | 2019-06-04 | 武汉南博网络科技有限公司 | A kind of pest and disease damage recognition methods and device |
| CN109840549B (en) * | 2019-01-07 | 2021-04-27 | 武汉南博网络科技有限公司 | Method and device for identifying plant diseases and insect pests |
| CN110689039A (en) * | 2019-08-19 | 2020-01-14 | 浙江工业大学 | Trunk texture identification method based on four-channel convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102760228B (en) | 2014-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102760228B (en) | Specimen-based automatic lepidoptera insect species identification method | |
| CN110543837B (en) | Visible light airport airplane detection method based on potential target point | |
| CN107563396B (en) | The construction method of protection screen intelligent identifying system in a kind of electric inspection process | |
| Díaz et al. | A semi-automatic method for quantification and classification of erythrocytes infected with malaria parasites in microscopic images | |
| CN105849274B (en) | Method and system for classification and identification of individual cells in microscopic images | |
| CN105574527B (en) | A kind of quick object detecting method based on local feature learning | |
| CN104573744B (en) | Fine granulation classification identifies and the part of object positions and feature extracting method | |
| CN108917593A (en) | A kind of intelligent measuring systems and method based on workpiece for measurement primitive configuration | |
| CN103345617A (en) | Method and system for recognizing traditional Chinese medicine | |
| CN107292314A (en) | A kind of lepidopterous insects species automatic identification method based on CNN | |
| Padilla et al. | Portable yellow spot disease identifier on sugarcane leaf via image processing using support vector machine | |
| Mirzaalian et al. | Automatic globally-optimal pictorial structures with random decision forest based likelihoods for cephalometric x-ray landmark detection | |
| Laga et al. | Image-based plant stornata phenotyping | |
| Borges et al. | Region growing for segmenting green microalgae images | |
| CN105138975B (en) | A kind of area of skin color of human body dividing method based on degree of depth conviction network | |
| CN106504262A (en) | A kind of small tiles intelligent locating method of multiple features fusion | |
| Ouyang et al. | The research of the strawberry disease identification based on image processing and pattern recognition | |
| CN107240112A (en) | Individual X Angular Point Extracting Methods under a kind of complex scene | |
| CN110599463A (en) | Tongue image detection and positioning algorithm based on lightweight cascade neural network | |
| CN103177266A (en) | Intelligent stock pest identification system | |
| Rahman et al. | Automatic detection of white blood cells from microscopic images for malignancy classification of acute lymphoblastic leukemia | |
| De | Automatic data extraction from 2D and 3D pie chart images | |
| CN111161295A (en) | Background stripping method for dish image | |
| EP3718046A1 (en) | Prostate cancer tissue image classification with deep learning | |
| Song et al. | Boundary-to-marker evidence-controlled segmentation and MDL-based contour inference for overlapping nuclei |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20140430 Termination date: 20200427 |