CN114898363A

CN114898363A - Identity recognition method and system based on egg shell image features

Info

Publication number: CN114898363A
Application number: CN202210584905.2A
Authority: CN
Inventors: 林宏建; 陈忠浩; 泮进明; 应义斌
Original assignee: Zhejiang University ZJU
Current assignee: Zhejiang University ZJU
Priority date: 2022-05-26
Filing date: 2022-05-26
Publication date: 2022-08-12
Anticipated expiration: 2042-05-26
Also published as: CN114898363B

Abstract

The invention discloses an identity recognition method based on egg shell image characteristics, which comprises the following steps: step 1, collecting an eggshell image of the surface of an egg, labeling the eggshell image with a label related to biological characteristics, and forming a sample set by the eggshell image and the label; step 2, constructing an egg identity recognition network for generating an egg identity code; step 3, training the egg identity recognition network by using the sample set, and obtaining an egg identity coding model after the training is finished; and 4, inputting the eggshell image of the egg into the egg identity coding model obtained in the step 3, and outputting the identity code for individual identification of the egg through calculation. The invention also discloses an identity recognition system based on the method. The method takes the image characteristics corresponding to the biological characteristics on the surface of the eggshell as the basis of identity coding, avoids the problem that the traditional physical label is counterfeited, and realizes the uniqueness and authenticity identification of the egg identity.

Description

Identity recognition method and system based on egg shell image features

Technical Field

The invention relates to the technical field of agricultural product production, in particular to an identification method and system based on egg shell image characteristics.

Background

Eggs are an indispensable important food source in daily life of people and are rich in amino acids, fatty acids and other nutrient substances. Particularly, high-end brand eggs such as organic eggs have higher nutritional value due to the high-quality feeding environment and feed formula, so that the high-end brand eggs are sold at higher price in the market. However, although such eggs are subject to industry certification or have brand marks of poultry egg enterprises for distinction, the eggs themselves have no obvious difference from common eggs, and it is difficult for human eyes to distinguish high-end brand eggs from common eggs. Due to the huge egg yield and the possible existing situation of unsupervised monitoring, for example, the simple machine of illegal household imitates brand egg package, packs common cheap eggs and even imitates the identification code of jet printing enterprise, thereby trying to deceive consumers in a false way.

In the prior art, an identification tag with specificity and anti-counterfeiting performance, such as a brand identification code, a digital number, an RFID electronic tag, a bar code, a two-dimensional code and other physical properties, is attached to an egg. The identity information of the eggs is stored by means of a third-party carrier, on one hand, the risk that the external physical labels are easy to lose and smear and pollute exists, on the other hand, the external physical labels are likely to be imitated by lawless persons, and the identity information of the invaded or tampered eggs is cracked and then is a label for sticking high-quality brand eggs to the cheap eggs.

Patent document CN109191461A discloses a native egg identification method and identification device based on machine vision technology, the identification method includes firstly taking a picture of an egg to be detected, calculating an egg shape index, a gray value information entropy and an eggshell surface impurity amount of the egg to be detected according to the picture, and then gradually comparing the egg shape index, the gray value information entropy and the eggshell surface impurity amount of the egg to be detected with a reference index range, a reference information entropy and a reference impurity amount respectively, thereby identifying the egg to be detected and judging whether the egg is a native egg. The method can only be used for identifying whether the eggs are the native eggs or not, and can not distinguish different types of native eggs, so that the method can not give technical inspiration for egg anti-counterfeiting.

Patent document CN114418265A discloses an agricultural product mark code identification system, including peasant household, consumer, APP or applet include user information registration module, personal center, sweep sign indicating number and image evolution identification technique, GPS location technique, online evaluation system and secret letter system, the peasant household maintains agricultural product basic information and sweeps the additional information that the sign indicating number was shot through APP, and the picture information that will additionally sweep the sign indicating number and shoot stores on the thing networking, after the consumer purchased agricultural product, through APP scanning identification two-dimensional code, utilize image evolution identification technique to discern agricultural product basic information and the additional information that the sign indicating number was shot of sweeping, realize agricultural product uniqueness and authenticity discernment. The method needs to combine the image of the current agricultural product with the attached two-dimensional code for identification, firstly, the appearance of part of the agricultural product can be changed continuously along with the time, so that all the agricultural products are not suitable for the method, secondly, the attached two-dimensional code can be counterfeited, and in conclusion, the method cannot solve the problem of agricultural product anti-counterfeiting.

Disclosure of Invention

In order to solve the problems, the invention provides an identification method based on image characteristics of egg shells, which is characterized in that image information of blunt ends of eggs is obtained, an independent identification code is endowed to each egg by using biological characteristics of the egg shell surfaces, and the biological characteristics of the egg shell surfaces of each egg are unique like human faces, so that the generated identification codes are difficult to counterfeit by people, and the biological characteristics can not change along with the change of time, thereby realizing the identification of uniqueness and authenticity of agricultural products.

An identity recognition method based on egg shell image features comprises the following steps:

step 1, collecting an eggshell image of the surface of an egg, labeling the eggshell image with a label related to biological characteristics, and forming a sample set by the eggshell image and the label;

step 2, an egg identity recognition network for generating an egg identity code is constructed, and the egg identity recognition network comprises a feature extraction module, a feature dimension reduction module and an identity editing module, wherein the feature extraction module comprises a multi-feature extractor for extracting various image features in an eggshell image, the feature dimension reduction module is used for performing dimension reduction processing on the various image features to obtain a low-dimensional image feature vector, and the identity coding module is used for coding the input image feature vector to output the identity code corresponding to the eggshell image;

step 3, training the egg identity recognition network by using the sample set, and obtaining an egg identity coding model after the training is finished;

and 4, inputting the eggshell image of the egg into the egg identity coding model obtained in the step 3, and outputting the identity code for individual identification of the egg through calculation.

Specifically, the biological features including spots, textures and bulges on the surface of the blunt end of the eggshell are fixed and do not change with time during production, and compared with the identity number automatically generated by the system, the safety is higher.

Preferably, before the eggshell image is input into the egg identification network, the eggshell image is preprocessed, including:

step 1.1, carrying out cluster analysis on pixel values of the eggshell image by adopting a K-means algorithm, removing patterns with similar pixel values in a non-eggshell region in the image, and obtaining a corresponding binary template image;

step 1.2, obtaining the circle center coordinate and radius size of the eggshell image by adopting a Hough circle detection algorithm;

step 1.3, performing mask processing on the eggshell image by using the binarized template image obtained in the step 1.1, and positioning and cutting the processed eggshell image according to the circle center coordinate and the radius size obtained in the step 1.2 to obtain an eggshell ROI image;

and step 1.4, performing size normalization on the eggshell ROI image, and inputting the eggshell ROI image into an egg identity recognition network.

Specifically, the egg identification network is formed by recombining a backbone network with VGGNet16 as feature extraction and a classification identification network of 1D _ CNN; the feature extraction capability of the original VGGNet16 network is combined with the feature dimension reduction capability of the 1D _ CNN network, so that the parameter number of the whole model is reduced, and the lightweight of the egg identification network is realized.

The invention also provides an identification system based on the egg shell image characteristics, which compares the eggshell image information provided by the user with the data in the database unit and performs anti-counterfeiting identification on the egg of the user according to the comparison result.

An identity recognition system is based on the identity recognition method based on the egg shell image features of the eggs, and comprises a server side and a client side, wherein the server side is used for recognizing the authenticity of the eggs, and the client side is used for acquiring images of the eggs to be recognized;

the server side comprises:

the production line shooting unit is used for acquiring image information of each egg blunt end on the production line;

the identity code generating unit is used for generating a corresponding identity code for the egg on the production line and generating a corresponding comparison identity code for the image information uploaded by the client based on the egg identity code model;

the database unit is used for storing the identity codes and the production information of the eggs;

the identity authentication unit analyzes and identifies the image information uploaded by the client and feeds back an identification result;

the client comprises:

the image acquisition unit is used for acquiring image information of the blunt end of the egg to be identified;

and the data transmission unit is used for carrying out data communication with the server side, and comprises uploading of image information and receiving of an identification result.

Specifically, production line shooting unit is including installing the camera in egg conveyer belt top to and arrange the light filling lamp circle at camera lens anterior segment, wherein, light filling lamp circle includes white light, red light, green light and blue light, thereby makes the image characteristic that the shooting was obtained more obvious.

Specifically, the identity authentication unit judges the similarity between the comparison identity code of the image information and the identity code stored in the database unit based on a pre-established judgment model.

Preferably, the judgment model judges based on a comparison result between the Euclidean distance calculation result and a threshold value:

if the identity code smaller than the threshold exists in the calculation result, outputting the egg as a true result;

and if the calculated result does not have the identity code smaller than the threshold value, outputting a false result of the egg.

Because the shooting environment of the production line is completely different from that of the client, including illumination, shooting angle and pixel level of the equipment, which all affect the extraction of the final image characteristics, the identity codes of two times of the same egg cannot be completely the same, and a threshold value is introduced to offset the influence of external factors.

Preferably, the threshold is obtained by performing statistics on probability distribution of Euclidean distances matching the same class and the different classes in the training process of the judgment model, so that the final similarity judgment result is more accurate.

Specifically, the identification result includes:

when the judgment result is true, feeding back the judgment result and the production information corresponding to the identity code;

and when the judgment result is false, only feeding back the judgment result and prompting the user to feed back the problem to the merchant.

Preferably, the database unit is further provided with a daily maintenance function, and the identity codes and the corresponding production information which exceed the quality guarantee date are deleted regularly according to the preset quality guarantee period of the eggs, so that a user is not guaranteed not to purchase expired products, and the storage pressure of the database unit can be reduced.

Compared with the prior art, the invention has the beneficial effects that:

(1) the identity is coded according to the biological characteristics of the surface of the eggshell, so that the danger that the physical label can be counterfeited is avoided.

(2) The traditional VGGNet16 network and the 1D _ CNN classification recognition network are recombined, so that the number of model parameters is reduced, and the problem of overhigh operation cost caused by excessive biological characteristics of eggshells is solved.

Drawings

Fig. 1 is a schematic flow chart of an identification method based on egg shell image characteristics according to the present invention;

fig. 2 is a schematic flowchart of eggshell image preprocessing provided in the present embodiment;

fig. 3 is a schematic structural diagram of an egg identification network provided in this embodiment;

fig. 4 is a schematic structural diagram of an identity recognition system provided in this embodiment;

fig. 5 is a schematic structural view of a production line shooting unit provided in the present embodiment;

in the figure, 1, a camera; 2. a semi-closed dark box; 3. an annular illumination source; 4. eggs; 5. egg conveyer belt.

Detailed Description

Nowadays, people pay more and more attention to the safety of agricultural products and food, and at present, the distinction between high-quality eggs and common eggs is generally distinguished by using two-dimensional codes or bar codes engraved on the surface of an eggshell, but the physical labels are very easy to counterfeit by people, so that some common eggs are mixed into the high-quality eggs to be sold by people.

Therefore, we propose a method for identifying eggs, as shown in fig. 1. An identity recognition method based on egg shell image features comprises the following steps:

step 1, collecting an eggshell image of an egg surface, labeling the eggshell image with spots, textures and protrusions on the surface of a blunt end of the eggshell, and forming the eggshell image and a corresponding label into a sample set, wherein in order to improve the accuracy of image recognition, the eggshell image needs to be preprocessed, as shown in fig. 2:

step 1.1, carrying out cluster analysis on the pixel values of the eggshell image by adopting a K-means algorithm to remove a background, and removing patterns which are not in the eggshell region but have similar pixel values in the image by adopting a Hough circle detection algorithm to obtain a binary template image corresponding to the eggshell region;

step 1.3, after the eggshell image is subjected to mask processing by using the binary template image obtained in the step 1.1, positioning and cutting the processed eggshell image according to the circle center coordinate and the radius size obtained in the step 1.2 to obtain an eggshell ROI image;

and step 1.4, after the size normalization is carried out on the eggshell ROI image, inputting the eggshell ROI image into an egg image feature extraction network.

Step 2, based on the framework of the VGGNet16 network, introducing a 1D _ CNN network to replace a full connection layer in the original network to construct an egg identification network, as shown in fig. 3.

The network comprises: the egg identity recognition network for generating the egg identity code is constructed and comprises a feature extraction module, a feature dimension reduction module and an identity editing module, wherein the feature extraction module comprises a multi-feature extractor and is used for extracting various image features in an eggshell image, the feature dimension reduction module is used for carrying out dimension reduction processing on the various image features to obtain a low-dimensional image feature vector, and the identity coding module is used for coding the input image feature vector to output the identity code corresponding to the eggshell image;

wherein, the structure of the 1D _ CNN network specifically includes: the first two layers are convolution layers, a Dropout layer is sequentially added after each convolution layer to reduce overfitting, and a BatchNorm layer is added to accelerate the convergence process of the neural network and improve the stability in the training process; and then adding a convolution layer, a Flatten layer and a Dense layer (namely a fully connected layer), and finally adding a Dropout layer, so that the neural network learns more stable characteristics and more effectively recombines the network.

Step 3, training the egg identity recognition network by using the sample set, and after the training is finished, obtaining an egg identity coding model: in the training process, an Adam optimizer is selected to iteratively update the weight parameters of the neural network based on training data, and the probability that an image output by a Softmax classifier is estimated to be an egg individual category is selected; the loss function used is the classification cross entropy; after training is finished, the Softmax classifier is removed.

The model performance evaluation can adopt the following indexes:

TP: true Positive, which is determined to be a Positive sample, and in fact is also a Positive sample;

TN: true Negative, determined as a Negative sample, in fact also a Negative sample;

FP: false Positive, judged as Positive, but in fact negative;

FN: false Negative, is judged as a Negative sample, but is in fact a positive sample.

The embodiment also provides an identity recognition system, as shown in fig. 4.

The system comprises a server side for identifying the authenticity of the egg and a client side for acquiring an image of the egg to be identified, wherein the server side is used for identifying the authenticity of the egg;

wherein, the server side includes:

the production line shooting unit has a specific structure as shown in fig. 5. The egg-shaped camera comprises a camera 1 arranged above an egg conveyor belt 5 and an annular illuminating light source 3 arranged at the front end of a camera lens, and in addition, a semi-closed camera bellows 2 capable of passing a single egg 4 is arranged on the conveyor belt to avoid the influence of external light;

the identity code generating unit generates a corresponding identity code for the egg on the production line and generates a corresponding comparison identity code for the image information uploaded by the client based on the egg identity code model;

the database unit is used for storing the identity code and the production information of the eggs, wherein the production information comprises a factory address, a production date, batch information and egg varieties;

the identity verification unit judges the identity codes in the database unit and the comparison identity codes by adopting a traditional judgment model, and feeds back a recognition result, wherein the judgment model adopts an Euclidean distance formula to calculate the similarity between the identity codes and the comparison identity codes, and compares the similarity result with a preset threshold value:

in the above formula, X, Y represents the identity code vectors of two eggshell images, ED represents the Euclidean distance, x ₁ ,x ₂ ,x ₃ ,…,x _n Representing points in a feature space X; y is ₁ ,y ₂ ,y ₃ ,…,y _n Representing points in a feature space Y;

if the calculation result does not have the identity code smaller than the threshold value, outputting a false result of the egg;

theoretically, the distance value of the Euclidean distance between the feature vectors extracted from the feature information of the eggshell image of the same egg is close to 0 (the similarity is close to 1).

Because the image feature information of the eggshell surface is complex and is interfered by various irresistible noises, and errors exist in the feature extraction result in the actual identification process, in order to effectively identify the eggshell biological features of different egg individuals, different distance thresholds suitable for egg type identification need to be set, and the threshold is obtained by counting the probability distribution of Euclidean distances matching the same type and different types in the process of training and judging a network model.

In the training process, the values of FAR and FRR are calculated, the transformation relationship between the FAR and FRR is statistically analyzed, and it is generally considered that when the values of FAR and FRR are equal, the distance threshold at this time is relatively adapted to the task of identifying the eggshell features, which is specifically expressed as:

the client comprises:

the image acquisition unit is used for acquiring image information of the blunt end of the egg to be distinguished;

and the data transmission unit is used for carrying out data communication with the server, and comprises the uploading of egg image information and the receiving of an identification result.

In addition, the identity recognition system is provided with a daily maintenance function, and the identity codes and the corresponding production information which exceed the quality guarantee date are deleted regularly according to the preset quality guarantee period of the eggs, so that a user is not guaranteed not to purchase expired products, and the storage pressure of the database unit can be reduced.

Claims

1. An identity recognition method based on egg shell image features is characterized by comprising the following steps:

2. The identification method according to claim 1, wherein the biological features include spots, textures and protrusions on the blunt end surface of the eggshell.

3. The identification method based on the egg shell image characteristics of the egg according to claim 1, wherein the egg shell image is preprocessed before being input into the egg identification network, and the preprocessing comprises:

4. An identification system based on egg shell image characteristics, the identification system is based on the identification method based on the egg shell image characteristics in any one of claims 1-3, and comprises a server side for identifying the authenticity of an egg and a client side for acquiring an image of the egg to be identified;

the server side comprises:

the client comprises:

5. The egg shell image feature-based identification system according to claim 4, wherein the production line shooting unit comprises a camera installed above the egg conveyor belt, and a supplementary lighting lamp ring arranged at the front end of the camera lens.

6. The identification system according to claim 4, wherein the identification unit performs similarity determination between the comparison identity code of the image information and the identity code stored in the database unit based on a pre-constructed determination model.

7. The identification system according to claim 6, wherein the judgment model is used for judging based on the comparison result of Euclidean distance calculation result and a threshold value:

8. The identification system according to claim 7, wherein the threshold is obtained by statistical distribution of probability of Euclidean distances matching same classes and different classes during training of the judgment model.

9. The identification system according to claim 4, wherein the identification result comprises: