CN112164120A - Dark spotted egg detection method based on convolutional neural network GoogLeNet model - Google Patents
Dark spotted egg detection method based on convolutional neural network GoogLeNet model Download PDFInfo
- Publication number
- CN112164120A CN112164120A CN202010994958.2A CN202010994958A CN112164120A CN 112164120 A CN112164120 A CN 112164120A CN 202010994958 A CN202010994958 A CN 202010994958A CN 112164120 A CN112164120 A CN 112164120A
- Authority
- CN
- China
- Prior art keywords
- egg
- dark
- model
- samples
- neural network
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 65
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 34
- 235000013601 eggs Nutrition 0.000 claims description 141
- 238000012549 training Methods 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 27
- 238000000034 method Methods 0.000 claims description 25
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 13
- 238000002474 experimental method Methods 0.000 claims description 12
- 230000005540 biological transmission Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 229910052742 iron Inorganic materials 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 210000003278 egg shell Anatomy 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 abstract description 5
- 230000004927 fusion Effects 0.000 abstract description 3
- 238000012795 verification Methods 0.000 abstract description 3
- 238000011160 research Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 102000002322 Egg Proteins Human genes 0.000 description 1
- 108010000912 Egg Proteins Proteins 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 210000002969 egg yolk Anatomy 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 244000005700 microbiome Species 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000005549 size reduction Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a dark spotted egg detection method based on a convolutional neural network GoogLeNet model, which comprises the steps of utilizing an inclusion module to repeatedly stack and construct a neural network framework, utilizing a multi-scale convolutional kernel to extract egg dark spot characteristics and carry out cascade fusion, designing an egg light-transmitting picture acquisition device in order to obtain the effectiveness of a sufficient picture sample verification model, obtaining 1200 dark spotted egg images and 8850 normal egg images in total, and selecting 1200 samples for network modeling. The problem of present dark spot egg detect mainly by artifical the completion, detect intensity of labour big, inefficiency is solved.
Description
Technical Field
The invention relates to the technical field of egg quality detection, in particular to a dark spotted egg detection method based on a convolutional neural network GoogLeNet model.
Background
The egg dark spots are dark spots formed by water in egg contents permeating and gathering on egg shells, the appearance of the dark spots is the expression of egg quality reduction, and the eggs with the dark spots are quick in water loss and freshness reduction, are more easily polluted by microorganisms and have adverse effects on egg storage performance.
At present, the detection of the dark spot eggs is mainly completed manually, and due to the fact that the coverage rate of the egg shells is large in difference, the positions of the egg shells are not fixed, the colors of partial dark spots are not obvious, the detection labor intensity is high, and the finding of a high-efficiency detection method has practical significance. The detection method combining machine vision and machine learning has the advantages of low use cost, high convenience and the like, and is widely researched in the aspect of egg quality detection, for example, the transparent image of the egg is collected by the Wangqiaohua method and the like, and the freshness of the egg is detected by extracting the color information in the egg or utilizing the morphological characteristics such as the area ratio of the yolk and the area ratio of the air chamber; tukang and the like mark the stains on the surface of the eggs by a threshold segmentation method to realize nondestructive detection of the stains on the eggs. The research makes a certain progress in the aspect of egg quality detection, but has limitations, for example, only dozens or hundreds of egg light transmission image samples are collected in experimental design, and a small sample easily causes an overfitting phenomenon; the number distribution of samples is not uniform, the characteristic distribution is unbalanced, for example, in a freshness detection experiment, the samples are concentrated in eggs at the second level, and the number of samples at the first level and the special level is far less than that of the samples at the second level; in addition, the research mostly uses traditional machine learning methods such as a BP neural network and a Support Vector Machine (SVM), the BP neural network needs to iteratively update network parameters for many times, and the SVM excessively depends on parameter adjustment; in addition, the extraction method based on color and morphological characteristics has the problems of more pretreatment steps, high interference and the like, and the existing egg quality detection method has low accuracy and has a large difference with actual production.
The deep learning Convolutional Neural Network (CNN) general convolution and downsampling operations are used for processing and classifying the characteristics of the multidimensional samples, so that the method has the advantages of processing multiple samples and large data, is widely applied to the field of image recognition and detection, and has few researches on egg quality detection by using the CNN.
Disclosure of Invention
The method aims to solve the problems that detection of dark spotted eggs in the prior art is mainly completed manually, the detection labor intensity is high, and the efficiency is low; the invention provides a dark spotted egg detection method based on a convolutional neural network GoogLeNet model.
In order to solve the above-mentioned existing technical problem, the invention adopts the following scheme:
a dark spotted egg detection method based on a convolutional neural network GoogLeNet model comprises the following steps that a, in order to obtain a light transmission image of an egg, an egg light transmission image acquisition system is set up, and an egg light transmission image sample is acquired;
b. the method comprises the following steps of (1) using a multi-angle acquisition mode for dark spot eggs, turning the eggs for 90 degrees after the eggs are subjected to image acquisition, repeatedly acquiring, not acquiring if the eggshells at a certain angle have no dark spots, and acquiring one image sample by using a normal egg; the collection is repeated every 1 day in the experiment, and the sampling period is about 20 days;
c. the color of the collected egg transmission light image is yellow close to red, the color of the dark spot of the egg is dark red, the color contrast of the two is not high, and the G component in the RGB space of the egg sample is enhanced by 4 times; then, the size of the G enhanced image is reduced to 1/8 of the original sample by utilizing an interpolation algorithm so as to meet the requirements of fast training and testing of a GoogleLeNet model;
d. in order to balance the quantity of the two types of samples, firstly, 1200 samples are randomly extracted from the normal egg samples, the two types of samples are processed in the step c, secondly, 900 samples are randomly selected from each type of egg samples according to the proportion of 1:3 to serve as model training samples, 300 samples are taken as test samples, and the labels of the dark spotted eggs and the normal eggs are coded by one-hot codes and are 0001 and 0010;
e. d, substituting the training sample and the label obtained in the step d into the input and output of a GoogleNet model for training, updating the weight by adopting a Stochastic Gradient Descent (SGD), and stopping training when the error or the iteration number reaches a threshold value; and finally, substituting the test sample into the trained network to obtain a detection result.
Further, step a egg printing opacity image acquisition system includes level crossing, ball integral light source, detected object egg, iron stand platform, CMOS industry camera, computer and software system detect ball integral light source bottom is placed the level crossing, detected object egg is placed at ball integral light source top, ball integral light source with the level crossing place in on the iron stand platform, CMOS industry camera is placed on the support of iron stand platform, CMOS industry camera passes through USB interface connection the computer, the computer is installed the image acquisition drive, realizes the software system detects.
Furthermore, the light source of the egg light-transmitting image acquisition system adopts an OPT-RID-150 sphere integral light source, and is combined with a plane mirror structure at the bottom of the light source, so that light rays from all directions of a spherical surface can be uniformly reflected, the light rays passing through the light holes at the top are uniform, and the illumination intensity and the transmission effect are excellent.
Further, the effective pixel of the industrial color CMOS camera is 2592 x 1944, 500 ten thousand pixels are 12mm lens, the collected ambient light intensity is 2-10Lux, the image is RGB color image, and the resolution is 1920 x 1440.
Further, the google lenet model used in the step c is composed of an input layer, a feature extraction layer, a full connection layer and an output layer.
Further, the feature extraction layer comprises a convolution layer and a down-sampling layer, the convolution layer is used for extracting features of the input image, the features extracted by different convolution kernels are different, the more the number of kernels of the convolution layer is, the more the extracted features are, and the down-sampling layer can reduce data processing amount and ensure operation speed.
Further, the google lenet model introduces multi-scale convolution to extract multi-scale local features by designing an inclusion module as a convolution layer, wherein the module comprises a plurality of 1 × 1, 3 × 3 and 5 × 5 convolution kernel branches.
Furthermore, a traditional CNN network basic module is adopted near the image input layer, 2 additional full-connection Softmax classifiers are added beside a main network by GoogLeNet, network model parameters are updated by the sum of loss function gradients of the main classifier and the branch classifiers, in the test process, the corresponding branch classifiers are removed, and only the main classifier is used for dark-spotted egg detection classification.
And step e, training and testing experiments, selecting the classification accuracy as an evaluation index, and optimizing by adopting an SGD algorithm.
Further, according to the test precision optimization principle, a convolutional network momentum parameter is set to be 0.7, the initial learning rate is (1e-4), the number of samples contained in a sample batch updated in each iteration of gradient descent is set to be 32, the epoch number is set to be 20, and a GoogLeNet model based on a convolutional neural network is established by utilizing the parameters to realize dark spot egg detection.
Compared with the prior art, the invention has the beneficial effects that:
(1) the problems that the detection of dark spotted eggs is mainly completed manually at present, the detection labor intensity is high, and the efficiency is low are solved;
(2) in order to obtain a light-transmitting image of an egg, an egg light-transmitting image acquisition device is set up and used for acquiring an egg light-transmitting image sample; the problems of too few samples, more pretreatment steps, low model precision and the like in the conventional egg quality detection research are solved, an egg image acquisition experiment is improved and designed, the number of samples is greatly increased in a multi-angle acquisition mode, and the number of samples of two types of eggs is balanced;
(3) the invention applies the convolutional neural network GoogLeNet model capable of automatically learning and extracting characteristics to egg dark spot detection, experimental division training and test samples are used for GoogLeNet model training, the detection accuracy of dark spot eggs is 98.19% in experimental results, and the CNN network GoogLeNet-based dark spot egg detection method is proved to have high detection accuracy without excessive preprocessing steps.
(4) The method utilizes the inclusion module to repeatedly stack and construct a neural network architecture, utilizes the multi-scale convolution kernel to extract the egg dark spot characteristics and carries out cascade fusion, has high feasibility and high detection precision in order to obtain the effectiveness of a sufficient picture sample verification model, and provides a new method for detecting the egg quality.
Drawings
FIG. 1 is a transparent image collecting device for eggs according to the present invention;
FIG. 2 is a flow chart of egg image processing according to the present invention;
FIG. 3 is a diagram of the structure of the GoogleLeNet model inclusion convolution module of the present invention;
FIG. 4 is a diagram of the GoogleLeNet model architecture of the present invention;
FIG. 5 is a flow chart of dark spotted egg detection based on the GoogleLeNet model according to the present invention;
fig. 6 is a graph of loss function values and accuracy generated by the google lenet model training of the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings and the detailed description, and it should be noted that any combination of the embodiments or technical features described below can be used to form a new embodiment without conflict.
(1) The characteristic of dark spots on the surface of the eggshell is selected, a set of egg light-transmitting image acquisition system is built, and a sample acquisition experiment is designed to obtain a sample.
(2) Normal eggs and dark spotted eggs are used as classification samples, and a method of combining machine vision with a Convolutional Neural Network (CNN) model is applied to egg dark spot detection.
(3) And (4) utilizing a GoogleLeNet model in CNN to learn the characteristics of the dark spots of the eggs, and establishing an egg dark spot detection model.
As shown in FIGS. 1 to 6, a method for detecting dark spotted eggs based on a convolutional neural network GoogLeNet model,
a. in order to obtain a light-transmitting image of an egg, an egg light-transmitting image acquisition system is set up, and an egg light-transmitting image sample is acquired;
b. the method comprises the following steps of (1) using a multi-angle acquisition mode for dark spot eggs, turning the eggs for 90 degrees after the eggs are subjected to image acquisition, repeatedly acquiring, not acquiring if the eggshells at a certain angle have no dark spots, and acquiring one image sample by using a normal egg; the collection is repeated every 1 day in the experiment, and the sampling period is about 20 days; the experiment obtains 1200 dark spotted egg images and 8850 normal egg images;
c. the color of the collected egg transmission light image is yellow close to red, the color of the dark spot of the egg is dark red, the color contrast of the two is not high, and the G component in the RGB space of the egg sample is enhanced by 4 times; then, the size of the G enhanced image is reduced to 1/8 of the original sample by utilizing an interpolation algorithm so as to meet the requirements of fast training and testing of a GoogleLeNet model;
d. in order to balance the quantity of the two types of samples, firstly, 1200 samples are randomly extracted from the normal egg samples, the two types of samples are processed in the step c, secondly, 900 samples are randomly selected from each type of egg samples according to the proportion of 1:3 to serve as model training samples, 300 samples are taken as test samples, and the labels of the dark spotted eggs and the normal eggs are coded by one-hot codes and are 0001 and 0010;
e. d, substituting the training sample and the label obtained in the step d into the input and output of a GoogleNet model for training, updating the weight by adopting a Stochastic Gradient Descent (SGD), and stopping training when the error or the iteration number reaches a threshold value; and finally, substituting the test sample into the trained network to obtain a detection result.
The egg light-transmitting image acquisition system comprises a plane mirror 0, a ball integral light source 1, an egg to be detected 2, an iron stand 3, a CMOS industrial camera 4, a computer 5 and a software system detection 6, wherein the plane mirror 0 is arranged at the bottom of the ball integral light source 1, the egg to be detected 2 is arranged at the top of the ball integral light source 1, the ball integral light source 1 and the plane mirror 0 are arranged on a platform of the iron stand 3, the CMOS industrial camera 4 is arranged on a support of the iron stand 3, the CMOS industrial camera 4 is connected with the computer 5 through a USB interface, and the computer 5 is provided with an image acquisition drive to realize the software system detection 6.
The egg light transmission image acquisition system is further improved in that the light source adopts an OPT-RID-150 sphere integral light source 1, and the structure of a plane mirror 0 at the bottom of the light source is combined, so that light rays from all directions of a spherical surface can be uniformly reflected, the light rays passing through the light holes at the top are uniform, and the illumination intensity and the transmission effect are excellent.
In a further improvement, the effective pixel of the industrial color CMOS camera is 2592 multiplied by 1944, 500 ten thousand pixels of a 12mm lens, the collected ambient light intensity is 2-10Lux, the image is an RGB color image, and the resolution is 1920 multiplied by 1440.
In an improved manner based on the above, the google lenet model used in the step c is composed of a picture input layer 21, a feature extraction layer 22, a full connection layer 24 and a classification result output layer 25, wherein the feature extraction layer 22 serves as a core of the network, and the greater the number of layers, the stronger the feature extraction capability.
Further, the feature extraction layer 22 includes a convolution layer and a down-sampling layer, the convolution layer is used for extracting features of the input image, the features extracted by different convolution kernels are different, the more the number of kernels of the convolution layer is, the more features are extracted, and the down-sampling layer can reduce data processing amount and ensure operation speed.
Meanwhile, the GoogleLeNet model introduces multi-scale convolution to extract multi-scale local features by designing an inclusion module as a convolution layer, wherein the module comprises a plurality of 1 × 1(12), 3 × 3(13) and 5 × 5(14) convolution kernel branches; the image input layer 11 adopts the basic modules of the traditional CNN network, namely a convolutional layer and a pooling layer, the characteristic of the middle layer is considered to have a certain degree of discrimination capability, the gradient disappearance problem in the optimization process of the random gradient descent algorithm caused by the fact that the network layer is too deep is also considered, 2 additional full-connection Softmax classifiers are added beside a trunk network by GoogLeNet, in the model optimization process, the sum of the loss function gradients of the trunk classifier and the branch classifiers is used for updating the parameters of the network model, in the test process, the corresponding branch classifiers are removed, and only the trunk classifier is used for carrying out dark spot egg detection classification.
In a further improvement, the difference algorithm used in the step c is a RESIZE function carried by the MATLAB2017b, so that image size reduction can be realized, and the google lenet model used in the step c is run on the MATLAB2017 b.
And (3) combining a GooglLeNet model-based dark spotted egg detection flow chart in FIG. 5:
obtaining 1200 dark spotted egg images and 8850 normal egg images in total, in order to balance the number of the two types of samples, firstly randomly extracting 1200 samples 32 from the normal egg samples, and 2400 samples 31 in total, and carrying out the egg image processing of the two types of samples in the graph 2; secondly, selecting the number of training sets and the number of testing sets, randomly dividing the training sets into 34 samples according to a ratio of 1:3, randomly taking 900 samples of each type of egg samples as model training samples, taking 300 samples of each type of egg samples as testing samples 35, and adopting one-hot coding for labels of dark spotted eggs and normal eggs, wherein the labels are 0001 and 0010; then substituting the training samples and the labels into the input and output of a GoogLeNet model for training, adopting a random gradient descent algorithm (SGD) optimization 36 for weight updating, and stopping training when the error or the iteration number reaches a threshold value; and finally, substituting the test sample into the trained network to obtain an egg dark spot detection result 37.
Loss function values and accuracy plots generated by training in conjunction with the google lenet model of fig. 6:
and step e, training and testing experiments, selecting the classification accuracy as an evaluation index, and optimizing by adopting an SGD algorithm.
The method is further improved in that according to the principle of optimal test precision, a convolutional network momentum parameter is set to be 0.7, the initial learning rate is (1e-4), the number of samples contained in a sample batch updated by each iteration of gradient descent is set to be 32, the epoch number is set to be 20, and a GoogLeNet model based on a convolutional neural network is established by utilizing the parameters to realize dark spot egg detection.
The test analyzes the change trend of the test accuracy and the loss function along with the increase of the iteration number in the model optimization process, and the experimental platform is MATLAB2017 b; CPU Intel (R) Xeon (R) X5650CPU @2.67GHz @2.67 GHz; the size of the memory is as follows: 48 GB. When the model is trained in the experiment, the model training condition is evaluated according to a Loss function (Loss) and an Accuracy (Accuracy) curve, and the network parameters are adjusted according to the model training condition.
The training loss function is in a descending trend in the training process, and the predicted loss deviation of the reaction model is gradually reduced through updating the gradient of the loss function in the optimization process. Meanwhile, as the iteration times are increased, the prediction accuracy of the model on the test set is in an overall rising trend. And (3) the training loss function is decreased, the prediction precision on the test set is increased, and the performance of the reaction model is optimized in the process of continuously iteratively updating parameters. When the number of iterations reaches 260, the convergence is substantial. The detection accuracy of the dark spotted eggs is 97.04%, and the detection method of the CNN network based dark spotted eggs is proved to have higher detection accuracy without excessive pretreatment steps.
In a further improvement, the SGD algorithm is a function carried by the GoogleLeNet tool box.
The method is further improved in that when the model is trained, the model training condition is evaluated according to a Loss function (Loss) and an Accuracy (Accuracy) curve, and the network parameters are adjusted according to the evaluation result.
(1) The problems that the detection of dark spotted eggs is mainly completed manually at present, the detection labor intensity is high, and the efficiency is low are solved;
(2) in order to obtain a light-transmitting image of an egg, an egg light-transmitting image acquisition device is set up and used for acquiring an egg light-transmitting image sample; the problems of too few samples, more pretreatment steps, low model precision and the like in the conventional egg quality detection research are solved, an egg image acquisition experiment is improved and designed, the number of samples is greatly increased in a multi-angle acquisition mode, and the number of samples of two types of eggs is balanced;
(3) the invention applies the convolutional neural network GoogLeNet model capable of automatically learning and extracting characteristics to egg dark spot detection, experimental division training and test samples are used for GoogLeNet model training, the detection accuracy of dark spot eggs is 98.19% in experimental results, and the CNN network GoogLeNet-based dark spot egg detection method is proved to have high detection accuracy without excessive preprocessing steps.
(4) The method utilizes the inclusion module to repeatedly stack and construct a neural network architecture, utilizes the multi-scale convolution kernel to extract the egg dark spot characteristics and carries out cascade fusion, has high feasibility and high detection precision in order to obtain the effectiveness of a sufficient picture sample verification model, and provides a new method for detecting the egg quality.
The above embodiments are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby, and any insubstantial changes and substitutions made by those skilled in the art based on the present invention are within the protection scope of the present invention.
Claims (10)
1. A dark spotted egg detection method based on a convolutional neural network GoogLeNet model is characterized by comprising the following steps:
a. in order to obtain a light-transmitting image of an egg, an egg light-transmitting image acquisition system is set up, and an egg light-transmitting image sample is acquired;
b. the method comprises the following steps of (1) using a multi-angle acquisition mode for dark spot eggs, turning the eggs for 90 degrees after the eggs are subjected to image acquisition, repeatedly acquiring, not acquiring if the eggshells at a certain angle have no dark spots, and acquiring one image sample by using a normal egg; the collection is repeated every 1 day in the experiment, and the sampling period is about 20 days;
c. the color of the collected egg transmission light image is yellow close to red, the color of the dark spot of the egg is dark red, the color contrast of the two is not high, and the G component in the RGB space of the egg sample is enhanced by 4 times; then, the size of the G enhanced image is reduced to 1/8 of the original sample by utilizing an interpolation algorithm so as to meet the requirements of fast training and testing of a GoogleLeNet model;
d. in order to balance the quantity of the two types of samples, firstly, 1200 samples are randomly extracted from the normal egg samples, the two types of samples are processed in the step c, secondly, 900 samples are randomly selected from each type of egg samples according to the proportion of 1:3 to serve as model training samples, 300 samples are taken as test samples, and the labels of the dark spotted eggs and the normal eggs are coded by one-hot codes and are 0001 and 0010;
e. and d, substituting the training sample and the label obtained in the step d into a GoogLeNet model for input and output to train, updating the weight by adopting a Stochastic Gradient Descent (SGD) algorithm, stopping training when the error or the iteration number reaches a threshold value, and finally substituting the test sample into a trained network to obtain a detection result.
2. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 1, characterized in that: step a egg printing opacity image acquisition system includes level crossing, ball integral light source, detected object egg, iron stand platform, CMOS industry camera, computer and software system detection ball integral light source bottom is placed the level crossing, detected object egg is placed at ball integral light source top, ball integral light source with the level crossing place in on the iron stand platform, CMOS industry camera is placed on the support of iron stand platform, CMOS industry camera passes through USB interface connection the computer, the computer is installed the image acquisition drive, realizes the software system detects.
3. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 2, characterized in that: the light source of the egg light-transmitting image acquisition system adopts an OPT-RID-150 sphere integral light source.
4. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 3, characterized in that: the effective pixel of the industrial color CMOS camera is 2592 multiplied by 1944, 500 ten thousand pixels are 12mm lenses, the collected ambient light intensity is 2-10Lux, the image is an RGB color image, and the resolution is 1920 multiplied by 1440.
5. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claims 1-4, characterized in that: and c, the GoogleLeNet model used in the step c is composed of an input layer, a feature extraction layer, a full connection layer and an output layer.
6. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 5, characterized in that: the feature extraction layer comprises a convolution layer and a down-sampling layer, the convolution layer is used for extracting features of an input image, the features extracted by different convolution kernels are different, the more the number of kernels of the convolution layer is, the more the extracted features are, and the down-sampling layer can reduce data processing amount and ensure operation speed.
7. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 6, characterized in that: the GoogLeNet model introduces multi-scale convolution to extract multi-scale local features by designing an inclusion module as a convolution layer, wherein the module comprises a plurality of 1 × 1, 3 × 3 and 5 × 5 convolution kernel branches.
8. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 7, characterized in that: the method comprises the following steps that a traditional CNN network basic module is adopted near an image input layer, 2 additional full-connection Softmax classifiers are added beside a main network by GoogLeNet, network model parameters are updated by the sum of loss function gradients of the main classifier and branch classifiers, the corresponding branch classifiers are removed in the testing process, and only the main classifier is used for detecting and grading the dark spotted eggs.
9. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 1, characterized in that: and e, training and testing experiments, selecting the classification accuracy as an evaluation index, and optimizing by adopting an SGD algorithm.
10. The dark spotted egg detection method based on the convolutional neural network GoogLeNet model as claimed in claim 9, characterized in that: according to the test precision optimization principle, setting a convolutional network momentum parameter to be 0.7, setting an initial learning rate to be (1e-4), setting the number of samples contained in a sample batch updated by each iteration of gradient descent to be 32, setting the number of epochs to be 20, and establishing a GoogLeNet model based on a convolutional neural network by using the parameters to realize dark spot egg detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010994958.2A CN112164120A (en) | 2020-09-21 | 2020-09-21 | Dark spotted egg detection method based on convolutional neural network GoogLeNet model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010994958.2A CN112164120A (en) | 2020-09-21 | 2020-09-21 | Dark spotted egg detection method based on convolutional neural network GoogLeNet model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112164120A true CN112164120A (en) | 2021-01-01 |
Family
ID=73863135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010994958.2A Pending CN112164120A (en) | 2020-09-21 | 2020-09-21 | Dark spotted egg detection method based on convolutional neural network GoogLeNet model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112164120A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114898363A (en) * | 2022-05-26 | 2022-08-12 | 浙江大学 | Identity recognition method and system based on egg shell image features |
-
2020
- 2020-09-21 CN CN202010994958.2A patent/CN112164120A/en active Pending
Non-Patent Citations (2)
| Title |
|---|
| MINLAN JIANG ET AL.: "Detecting Dark Spot Eggs Based on CNN GoogLeNet Model", 《 INTERNATIONAL CONFERENCE ON SIMULATION TOOLS AND TECHNIQUES》, vol. 370, pages 116 * |
| 李飞: "机器学习在鸡蛋产量预测与品质检测中的应用", 《中国优秀硕士学位论文全文数据库 (农业科技辑)》, no. 2, pages 050 - 264 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114898363A (en) * | 2022-05-26 | 2022-08-12 | 浙江大学 | Identity recognition method and system based on egg shell image features |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108304873B (en) | Target detection method and system based on high-resolution optical satellite remote sensing image | |
| CN109784204B (en) | Method for identifying and extracting main fruit stalks of stacked cluster fruits for parallel robot | |
| CN110276386A (en) | A kind of apple grading method and system based on machine vision | |
| CN109509187B (en) | Efficient inspection algorithm for small defects in large-resolution cloth images | |
| CN113591766B (en) | Multi-source remote sensing tree species identification method for unmanned aerial vehicle | |
| CN106557778A (en) | Generic object detection method and device, data processing equipment and terminal device | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| Zuo et al. | HF-FCN: Hierarchically fused fully convolutional network for robust building extraction | |
| CN109948696A (en) | A kind of multilingual scene character recognition method and system | |
| CN111178177A (en) | Cucumber disease identification method based on convolutional neural network | |
| CN112766155A (en) | Deep learning-based mariculture area extraction method | |
| CN110929795A (en) | Method for quickly identifying and positioning welding spot of high-speed wire welding machine | |
| CN112464983A (en) | Small sample learning method for apple tree leaf disease image classification | |
| Ünal et al. | Classification of hazelnut kernels with deep learning | |
| Yang et al. | Instance segmentation and classification method for plant leaf images based on ISC-MRCNN and APS-DCCNN | |
| CN112132803A (en) | Egg freshness detection method based on convolutional neural network | |
| CN117011260A (en) | Automatic chip appearance defect detection method, electronic equipment and storage medium | |
| CN114898097A (en) | Image recognition method and system | |
| Kumar et al. | A deep learning paradigm for detection of harmful algal blooms | |
| CN115294033A (en) | Tire belt layer difference level and misalignment defect detection method based on semantic segmentation network | |
| CN110929787B (en) | Apple objective grading system based on image | |
| Jana et al. | Detection of rotten fruits and vegetables using deep learning | |
| Han et al. | Deep learning based approach for automated characterization of large marine microplastic particles | |
| CN115731198A (en) | Intelligent detection system for leather surface defects | |
| CN112164120A (en) | Dark spotted egg detection method based on convolutional neural network GoogLeNet model |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210101 |