CN112926648B - Method and device for detecting abnormality of tobacco leaf tip in tobacco leaf baking process - Google Patents
Method and device for detecting abnormality of tobacco leaf tip in tobacco leaf baking process Download PDFInfo
- Publication number
- CN112926648B CN112926648B CN202110205229.9A CN202110205229A CN112926648B CN 112926648 B CN112926648 B CN 112926648B CN 202110205229 A CN202110205229 A CN 202110205229A CN 112926648 B CN112926648 B CN 112926648B
- Authority
- CN
- China
- Prior art keywords
- tobacco leaf
- tip
- tobacco
- image
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a device for detecting abnormality of a tobacco leaf tip part in a tobacco leaf baking process, which relate to the technical field of tobacco leaf baking and comprise the following steps: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera; inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram; determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram; and inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolutional neural network, and judging whether the tobacco leaf tip is abnormal or not. According to the tobacco leaf detection method, the convolutional neural network is used for predicting the thermodynamic diagram, then the position of the tobacco tip is located in a mode of finding a local maximum value in the thermodynamic diagram, the thermodynamic diagram is used as a label, each pixel on the diagram can provide supervision information when the network is trained, and the tobacco leaf detection accuracy is improved.
Description
Technical Field
The invention belongs to the technical field of tobacco leaf baking, and particularly relates to a method and a device for detecting abnormal tobacco leaf tip parts in a tobacco baking process based on a convolutional neural network.
Background
The convolutional neural network is a feedforward neural network which comprises convolutional calculation and has a deep structure, and is one of representative algorithms of deep learning. Convolutional neural networks have a characteristic learning ability, and can perform translation invariant classification on input information according to a hierarchical structure thereof, and are also called translation invariant artificial neural networks. The traditional convolution neural network is composed of convolution layers and full connection layers and is mostly suitable for image classification tasks, and the full convolution neural network only uses the convolution layers and obtains better effects in the field of image segmentation in recent years.
Before the neural network is used, the neural network must be trained by processed data (the data comprises two parts of input data and label data), and output data can be predicted according to the input data after training is finished. For a convolutional neural network for classification, the input data is usually a picture, and the label data is usually a set of vectors (the vector dimension corresponds to the number of classes to be predicted); for a full convolution neural network, the input data and the tag data are usually both one picture.
The first step of the cigarette making process is to bake the collected fresh tobacco leaves. The traditional tobacco leaf baking process is to put the collected tobacco leaves into a special baking room, the expert sets a preset baking curve on baking room equipment, and then the baking room controls the temperature according to the curve preset by the expert, so that the tobacco leaves in the baking room are in proper temperature and humidity states.
The tobacco leaves can generate obvious morphological changes in the baking process, and can be divided into three major stages of yellowing, fixing color and drying stem according to the color, crease and curling degrees of the tobacco leaves; wherein each large stage can be further divided into smaller stages (initial yellowing stage, middle yellowing stage, later yellowing stage, earlier fixing stage, middle fixing stage, later fixing stage, early stem stage, middle stem stage, and later stem stage); the temperature curve that the expert predetermines on "roast room equipment" just corresponds with every little stage one-to-one that the tobacco leaf was located, for example "yellow initial stage" sets up to 32 degrees, "stem later stage" sets up to 68 degrees, by the aforesaid, the stage division of tobacco leaf is obtained by the tobacco leaf "colour", "fold", "curling" degree, and these characteristics all can calculate through the tobacco leaf picture that the camera was shot, consequently, combine camera and automation technology to toast the novel intelligent roast room of inoculating and come out.
The tip of the tobacco leaf is the thinnest part in the whole tobacco leaf and is also the part which can show the change most quickly. Therefore, the tip of the tobacco leaf is selected as the first index of the temperature and humidity change for judgment, and the trend of the change of the whole tobacco leaf to be caused can be found in advance. In the yellowing stage of tobacco leaf baking, if the internal temperature and humidity of the baking room are too high, the tip part of the tobacco leaf and other parts of the tobacco leaf will be yellow and black. The blackening of the tip of the tobacco leaves can affect the quality of the tobacco leaves and reduce the purchase price.
Disclosure of Invention
Aiming at the problem that the tip part of the tobacco leaf is easy to yellow and blacken in advance in the yellowing stage, the invention provides a method and a device for detecting the abnormality of the tip part of the tobacco leaf in the tobacco leaf baking process, so that the abnormality detection of the tip part of the tobacco leaf based on image analysis in the tobacco leaf baking process is realized, and the quality of the tobacco leaf is effectively guaranteed. Therefore, the present invention adopts the following technical solutions.
In a first aspect, the invention provides a method for detecting abnormality of a tobacco leaf tip in a tobacco leaf baking process, which comprises the following steps:
s1: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera;
s2: inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram;
s3: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
s4: inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain probability values of normal tobacco leaf tip and abnormal tobacco leaf tip;
s5: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
Further, the step of constructing the trained tobacco tip positioning convolutional neural network in the step S2 is as follows:
the first step is as follows: acquiring a blade tip positioning image sample set through a camera;
the second step is that: preprocessing the blade tip positioning image sample set, interpolating the position with the cigarette tip by taking a Gaussian function as a template, and manufacturing a blade tip positioning image sample set label;
the third step: constructing a tobacco leaf tip positioning convolution neural network;
the fourth step: and performing iterative training on the preprocessed blade tip positioning image sample set by adopting the convolutional neural network in the previous step to obtain the trained tobacco tip positioning convolutional neural network.
Further, the manufacturing steps of the label are as follows:
firstly, creating a blank image set which is in one-to-one correspondence with images in the blade tip positioning image sample set, wherein the size of the images in the blank image set is consistent with that of the images in the blade tip positioning image sample set;
secondly, marking the position of the tip of the tobacco leaf in each image in the blade tip positioning image sample set;
thirdly, label making, namely determining the position corresponding to the tobacco leaf tip part position on the corresponding blank image according to the tobacco leaf tip part position, and performing interpolation by taking a two-dimensional Gaussian function as a template according to the position corresponding to the tobacco leaf tip part position in the blank image, wherein the central value of the two-dimensional Gaussian function is the position of the corresponding tobacco leaf tip part in the blank image;
and finally, making a corresponding label for each image in the blank image set.
Further, when the label is manufactured, when at least 2 two-dimensional Gaussian functions have overlapping areas in the action range, taking the maximum value on the corresponding Gaussian function as the position of the corresponding tobacco leaf tip in the blank image in the overlapping areas.
Further, the tobacco leaf tip positioning convolutional neural network is a stacked hourglass network formed by 2 densely connected hourglass structure networks.
Further, in the tobacco leaf thermodynamic diagram, the value of each pixel point represents the probability that the pixel point has a tobacco tip part, when the value of the pixel point is greater than or equal to the value of the pixel point in a 15 × 15 area taking the pixel point as the center, the value of the pixel point is the local maximum value, the pixel point corresponding to the local maximum value is the tobacco leaf tip part position, the tobacco leaf tip part position in the thermodynamic diagram is taken as the center, and a tobacco leaf tip part picture is intercepted from a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram.
Further, the tobacco tip abnormality classification convolutional neural network construction step in the step S4 is as follows:
the first step is as follows: acquiring a tobacco leaf image sample set through a camera, and marking and intercepting a leaf tip part area in the tobacco leaf image set to obtain a leaf tip abnormal classification image sample set;
the second step is that: preprocessing the abnormal leaf tip classified image sample set, manufacturing an abnormal leaf tip classified image sample set label, classifying the abnormal leaf tip classified image sample set according to whether the tip of the tobacco leaves in the abnormal leaf tip classified image sample set is blackened, wherein the label is that the tip of the tobacco leaves is abnormal, and otherwise, the tip of the tobacco leaves is normal;
the third step: constructing a tobacco leaf abnormal classification convolutional neural network;
the fourth step: and performing iterative training on the preprocessed leaf tip abnormity classification image sample set by adopting the convolution neural network in the last step to obtain the trained tobacco leaf tip abnormity classification convolution neural network.
Further, the tobacco leaf abnormity classification convolutional neural network is formed by sequentially connecting 1 average pooling layer of 16 convolutional layers with convolution kernel size of 3 x3 and 1 full-connection layer, wherein the full-connection layer outputs 2 values, and the values are correspondingly probability values of normal tobacco leaf tip and abnormal tobacco leaf tip.
Further, in the step S5, when the probability value of the normal tip portion of the tobacco leaf is greater than the probability value of the abnormal tip portion of the tobacco leaf, outputting the recognition result of the tip portion state of the tobacco leaf as that the tip portion of the tobacco leaf is normal; and when the probability value of the normal tip part of the tobacco leaves is less than or equal to the probability value of the abnormal tip part of the tobacco leaves, outputting the state recognition result of the tip part of the tobacco leaves as the abnormal tip part of the tobacco leaves.
In a second aspect, the present application provides a device for detecting an abnormality of a tip portion of a tobacco leaf during a tobacco leaf curing process, comprising:
a tobacco leaf image acquisition module: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera;
a tobacco leaf thermodynamic diagram acquisition module: inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram;
a tobacco tip identification module: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
the tobacco leaf tip abnormal probability output module: inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain probability values of normal tobacco leaf tip and abnormal tobacco leaf tip;
the tobacco leaf tip abnormity identification output module: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
In a third aspect, the present application provides a terminal, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for detecting the abnormality of the tip of the tobacco during the tobacco flue-curing process when executing the program.
In a fourth aspect, the present application provides a computer storage medium having a computer program stored thereon, which when executed by a processor, implements the steps of a method for detecting an abnormality of a tip portion of a tobacco leaf during a tobacco leaf curing process.
The invention has the beneficial effects that: the invention provides a method for detecting the abnormality of the tobacco leaf tip in the tobacco leaf baking process, which mainly aims at the tobacco leaf state identification in the tobacco leaf yellowing stage, adopts a target detection algorithm based on a convolutional neural network, and positions the tobacco leaf tip position by enabling the convolutional neural network to predict a thermodynamic diagram and then finding a local maximum value in the thermodynamic diagram, thereby realizing the high-precision automatic tobacco leaf tip identification and extraction; meanwhile, because the thermodynamic diagram is used as a label, each pixel on the diagram can provide supervision information when the network is trained, a large amount of supervision information can enable the network to be effectively converged under a smaller training sample set, and the accuracy of tobacco leaf detection is improved.
Drawings
FIG. 1 is a schematic flow chart of a method for detecting abnormality of a tobacco leaf tip part in a tobacco leaf curing process according to the present invention;
FIG. 2 is a schematic diagram of a label of a leaf tip positioning image sample set in a method for detecting abnormality of a leaf tip in a tobacco leaf curing process according to the present invention;
fig. 3 is a schematic diagram of a sample set label of the tip positioning image of the method for detecting abnormality of the tip portion of the tobacco during the tobacco flue-curing process, which is provided by the invention, being shown in an original drawing;
fig. 4 is a schematic structural diagram of the convolutional neural network in step S2 of the method for detecting abnormality of the tip portion of the tobacco during the tobacco flue-curing process according to the present invention;
fig. 5 is a schematic structural diagram of the convolutional neural network in step S4 of the method for detecting abnormality of the tip portion of the tobacco during the tobacco flue-curing process according to the present invention;
fig. 6 is a block diagram of a device for detecting abnormality of the tip portion of tobacco leaves in the tobacco leaf curing process.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
In order to overcome the problems in the prior art, the invention provides a method and a device for detecting the abnormality of the tip part of the tobacco leaf in the tobacco leaf baking process.
Referring to fig. 1, an embodiment of a method for detecting an abnormality of a tobacco leaf tip portion in a tobacco leaf curing process according to the present invention is that, as shown in fig. 1, the method for detecting an abnormality of a tobacco leaf tip portion in a tobacco leaf curing process specifically includes the steps of:
s1: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera;
s2: inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram;
s3: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
s4: inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain probability values of normal tobacco leaf tip and abnormal tobacco leaf tip;
s5: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
In step S1, specifically, in the tobacco leaf baking process, the camera is used to acquire and obtain the tobacco leaf image in the tobacco leaf baking process in a preset acquisition period, where the preset acquisition period is, for example, 1min, 5min, 8min, 10min, 15min, 16min, 20min, and the like, and the numerical values are only used for illustration and are not limited specifically.
In step S2, inputting the tobacco leaf image into the trained tobacco leaf tip positioning convolutional neural network to obtain a tobacco leaf thermodynamic diagram, firstly training the tobacco leaf tip positioning convolutional neural network to obtain the trained tobacco leaf tip positioning convolutional neural network, wherein the concrete tobacco leaf tip positioning convolutional neural network construction set training step is as follows:
the first step is as follows: acquiring a blade tip positioning image sample set through a camera;
in the embodiment, the camera is used for acquiring the tobacco leaf images with various angles, different sizes, different shapes and different states under various illumination conditions, including the tobacco leaf images at the early yellowing stage, the middle yellowing stage and the later yellowing stage as the leaf tip positioning image sample set, and 1000 cameras are used for acquiring the tobacco leaf images with various angles, different sizes, different shapes and different states of the tobacco leaf yellowing under various illumination conditions as the leaf tip positioning image sample set.
The second step is that: preprocessing the blade tip positioning image sample set, interpolating the position with the cigarette tip by taking a Gaussian function as a template, and manufacturing a blade tip positioning image sample set label;
the preprocessing comprises zooming, cutting, rotating and manual marking, most of interference factors of non-key areas can be removed by preprocessing the picture, the picture is corrected, the phenomenon that subsequent tobacco leaf tip identification is influenced by the blurring or distortion of the picture is avoided, the subsequent identification result can be verified and compared by manually marking the picture in the training set, and therefore network parameters are optimized.
Specifically, the pretreatment step may include:
(1) and (3) cutting, namely, scaling the images in the collected blade tip positioning image sample set to a size suitable for neural network input, such as 512 × 512, 32 × 32, 64 × 64 and the like, and using the scaled images as input for neural network training, wherein numerical values are only used for illustration and are not limited specifically.
(2) And rotating or overturning the images in the blade tip positioning image sample set to form an amplification training sample. The rotation angle randomly takes a value in the range of 0-20 degrees, the overturning direction randomly takes horizontal overturning or longitudinal overturning, the sample diversity is increased, and the generalization performance of the convolutional neural network training model is improved.
(3) And performing HSV color space conversion on the images in the blade tip positioning image sample set from the RGB images to obtain HSV color spaces, enhancing different coefficients of the converted images of the H channel, the S channel and the V channel, wherein the enhancing coefficient takes a value of 0.7-1.4, and then converting the enhanced HSV color space images back to the RGB images, thereby realizing color enhancement of the training samples.
Specifically, the manufacturing steps of the leaf apex positioning image sample set label are as follows:
further, the manufacturing steps of the label are as follows:
first, a blank image set corresponding to images in the tip positioning image sample set one by one is created, for example, if the size of the images in the tip positioning image sample set is 512 × 512, blank images with the same size are created, and the images in the tip positioning image sample set correspond to the images in the blank image set one by one.
Secondly, marking the position of the tip of the tobacco leaf in each image in the blade tip positioning image sample set; the cigarette tips are marked by manually using a marking tool in a point marking mode, namely, if the cigarette tips are arranged at a certain position on the collected image, a point is marked at the position by using the marking tool.
And thirdly, label making, namely determining the position corresponding to the tobacco tip part position on the corresponding blank image according to the tobacco tip part position, carrying out interpolation by taking a two-dimensional Gaussian function as a template according to the position corresponding to the tobacco tip part position in the blank image, wherein the central value of the two-dimensional Gaussian function is the position of the corresponding tobacco tip part in the blank image.
The gaussian formula used is:
wherein (x, y) represents a two-dimensional coordinate value; mu.s1，μ2，σ1，σ2And ρ is a parameter of a two-dimensional Gaussian function, specifically, μ is taken in this example1＝0，μ2＝0，σ1＝3，σ2＝3，ρ＝0。
And when the action ranges of at least 2 two-dimensional Gaussian functions have overlapping areas, taking the maximum value on the corresponding Gaussian function as the position of the corresponding tobacco tip in the blank image in the overlapping areas. And when the tobacco leaf tip part position in the image of the blade tip positioning image sample set can find the corresponding Gaussian function interpolation value, namely the tobacco leaf tip part position, on the blank image corresponding to the tobacco leaf tip part position, and the label of the image is manufactured.
Referring to fig. 2 and 3, schematic diagrams of tip positioning image sample set labels shown in an original drawing are shown, the center of a gaussian function corresponds to the position of a cigarette tip, the corresponding function value is the largest, the farther the distance from the center is, the smaller the function value is, and the probability distribution situation at the position of the cigarette tip is described in this way. In the label, even if belonging to the same class (e.g., the class of the cigarette tip), the pixel values are not identical everywhere but distributed according to a two-dimensional gaussian function. As shown in the following figures: white pixels all represent the same class (tip class). However, some pixels have larger values (higher opacity) and some pixels have smaller values (higher transparency), as shown in fig. 3.
And finally, making a corresponding label for each image in the blank image set.
The third step: constructing a tobacco leaf tip positioning convolution neural network:
the tobacco leaf tip positioning convolutional neural network in the embodiment is a stacked hourglass network formed by 2 densely connected hourglass structure networks.
Referring to fig. 4, each rectangular box represents a neural network layer, the network layers appearing in the model include a batch normalization layer (BnReluConv, which makes the network easier to train), a Residual layer (Residual, which improves the network feature extraction capability and simultaneously prevents the network from overfitting), a maximum pooling layer (MaxPool2d, which can reduce the size of input data to half of the input data and remove redundant image information), a Hourglass structure module (Hourglass, which obtains location information), a 1x1 convolution layer (1x1conv, which compresses the input information and improves the efficiency of the network), and the whole cigarette tip location model is formed by connecting the network layers according to arrows shown in the figure, wherein a plus sign represents that information of different network layers is fused, and the fusion mode is that numerical values of corresponding locations on output data of the network layers are directly added. The model has outputs out 0 and out 1 in the middle and final layers, and when network training is performed, the parameters of the network are trained by using out 0 and out 1 at the same time, and after the network training is completed, only the output of out 1 is used as the result of network inference.
The specific structure of the tobacco leaf tip positioning convolution neural network is as follows:
preprocessing the images in the preprocessed blade tip positioning image sample set sequentially through a batch normalization layer (BnReluconv), a Residual layer (Residual), a maximum pooling layer (Maxpool2d), a Residual layer (Residual) and a Residual layer (Residual) to obtain a preliminary image characteristic f 1;
inputting f1 into a first Hourglass structure module (Hourglass), wherein the Hourglass structure module firstly carries out four times of downsampling and retains a feature map of each downsampling; then, taking the feature map with the minimum size as initial input, performing up-sampling for four times, and connecting the feature map with the size corresponding to the down-sampling each time, thereby finally obtaining an output feature map f 2;
processing f2 sequentially through a Residual layer (Residual), a Residual layer (Residual) and a batch normalization layer (BnReluconv) to obtain a characteristic diagram f3, and processing f3 through a 1x1 convolution layer (1x 1conv) to obtain a thermodynamic diagram output by a first network and marking as output [0 ];
processing f3 through a 1x1 convolution layer (1x 1conv) to obtain a feature map f4, processing output [0] through a 1x1 convolution layer (1x 1conv) to obtain a feature map f5, and summing three feature maps of f1, f4 and f5 according to numerical values of corresponding positions to obtain a feature map f 6;
inputting f6 into a second Hourglass structure module (Hourglass), wherein the Hourglass structure module firstly performs four downsampling operations and retains a feature map of each downsampling operation; then, taking the feature map with the minimum size as initial input, performing up-sampling for four times, and connecting the feature map with the size corresponding to the down-sampling each time, thereby finally obtaining an output feature map f 7;
processing f7 sequentially through a Residual layer (Residual), a Residual layer (Residual) and a batch normalization layer (BnReluconv) to obtain a characteristic map f8, and processing f8 through a 1x1 convolution layer (1 × 1conv) to obtain a thermodynamic output [1] output of the second network.
The fourth step: and performing iterative training on the preprocessed blade tip positioning image sample set by adopting the convolutional neural network in the previous step to obtain the trained tobacco tip positioning convolutional neural network.
The specific iterative training process is as follows: the preprocessed apex positioning image sample set is used for training a full convolution neural network, a loss function in the network training process adopts a mean square error function MSE, and the formula is as follows:
y in the formulai ylabelRespectively an output obtained after an input picture is input into the network model and a label picture, wherein n is the logarithm of the pictures in the training data (one input picture and one corresponding label picture form a pair); m is the total number of pixels of the label picture,
the pixel values at the corresponding positions of the two pictures are subtracted by a square value and are summed. In the training process, when the current iteration error of the last two times tends to be unchanged, the network confirms that the iteration error is converged, the training is stopped, and the trained tobacco tip positioning convolution neural network is obtained.
In the iterative training process in this embodiment, when the error value MSE is less than 10-7 and the ratio of the error values of the previous time and the next time is greater than 99.999%, it is determined that the network has converged and the training is stopped.
And after the trained tobacco tip positioning convolution neural network is obtained, the tobacco image collected in the step S1 is input into the trained tobacco tip positioning convolution neural network to output the thermodynamic diagram with the same size.
Step S3: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
on the thermodynamic diagram, the value of each pixel point represents the prediction probability of the cigarette tip part at the position, the larger the value is, the higher the probability of the cigarette tip is, on the thermodynamic diagram, a local maximum value is searched, and the position of the maximum value is the prediction position of the cigarette tip.
In order to determine the position of the tip of the tobacco in one area through one local maximum, the size of the area for searching the local maximum is controlled to ensure that the number of the tip of the tobacco which is found is only one and corresponds to the position where the tip of the tobacco really exists.
Specifically, in the tobacco leaf thermodynamic diagram, the value of each pixel point represents the probability that the pixel point has a tobacco tip part, when the value of the pixel point is greater than or equal to the value of the pixel point in a 15 × 15 region with the pixel point as the center, the value of the pixel point is the local maximum value, the pixel point corresponding to the local maximum value is the tobacco leaf tip part position, a tobacco leaf tip part picture is cut out from a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram by taking the tobacco leaf tip part position in the thermodynamic diagram as the center, the size of the cut-out picture is based on that the proportion of the area containing the tobacco leaves in the picture to the total area of the picture is greater than 30%, and the size of the cut-out tobacco leaf tip part image in the embodiment is 32 × 32.
Step S4: and inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain the probability value of the normal tobacco leaf tip and the abnormal tobacco leaf tip.
The first step is as follows: acquiring a tobacco leaf image sample set through a camera, and marking and intercepting a leaf tip part area in the tobacco leaf image set to obtain a leaf tip abnormal classification image sample set;
the embodiment collects tobacco leaf images with various angles, different sizes, different shapes and different states under various illumination conditions through the camera, wherein the tobacco leaf images comprise tobacco leaf images at the early yellowing stage, the middle yellowing stage and the later yellowing stage. Manually identifying and cutting a leaf tip area in a tobacco leaf image in the leaf tip image obtaining process, manually marking a picture in a tobacco leaf image sample set in a marking point mode by using a marking tool, specifically marking a point at a certain position on the tobacco leaf image, cutting a 32x32 tobacco leaf tip picture by using a script program with the marking position as a center, and collecting all obtained tobacco leaf tip pictures together to form a leaf tip abnormal classification image sample set. In the embodiment, 2000 cameras are used for collecting tobacco leaf images with various angles, different sizes, different shapes and different states of tobacco leaf yellowing under various illumination conditions, and the leaf tip images are captured to obtain a leaf tip abnormal classification image sample set.
The second step is that: preprocessing the abnormal leaf tip classified image sample set to manufacture a abnormal leaf tip classified image sample set label;
the preprocessing comprises zooming, cutting, rotating and manual marking, most of interference factors in non-key areas can be removed by preprocessing the pictures, the pictures are corrected, the influence of blurring or distortion of the pictures on subsequent abnormal tobacco leaf tip classification is avoided, the training set pictures are manually marked, the subsequent abnormal tobacco leaf tip classification results can be verified and compared, and therefore network parameters are optimized.
Specifically, the pretreatment step may include:
(1) and (3) cutting and scaling, namely scaling the images in the collected abnormal tip classification image sample set to a size suitable for neural network input, such as 32 × 32 and 64 × 64, and using the scaled images as the input of neural network training, wherein numerical values are only used for illustration and are not specifically limited.
(2) And rotating or overturning the images in the abnormal tip classified image sample set to form an amplification training sample. The rotation angle randomly takes a value in the range of 0-20 degrees, the overturning direction randomly takes horizontal overturning or longitudinal overturning, the sample diversity is increased, and the generalization performance of the convolutional neural network training model is improved.
(3) And performing HSV color space conversion on the images in the tobacco abnormal classification image sample set from the RGB images to obtain HSV color spaces, enhancing different coefficients of the converted images of the H channel, the S channel and the V channel, wherein the enhancing coefficient takes a value of 0.7-1.4, and then converting the enhanced HSV color space images back to the RGB images, thereby realizing color enhancement of the training samples.
Specifically, a leaf tip abnormal classification image sample set label is manufactured, and a 'tobacco tip picture' in the intercepted leaf tip abnormal classification image sample set is classified according to whether the tip of the tobacco leaf is blackened or not, wherein the label is that the tip of the tobacco leaf is abnormal, otherwise, the tip of the tobacco leaf is normal;
the third step: constructing a tobacco leaf abnormal classification convolutional neural network;
specifically, in this embodiment, a resnet18 model is used to construct a tobacco leaf anomaly classification convolutional neural network, and specifically, the convolutional neural network is formed by sequentially connecting 1 average pooling layer and 1 full-link layer of 16 convolutional layers with a convolutional kernel size of 3 × 3.
The leaf anomaly classification convolutional neural network structure is shown in FIG. 5, each box represents a neural network layer, except that an "FC 10" represents a full-link layer and an "avg pool" represents an average pooling layer, the rest layers are convolutional layers, the convolutional cores have the size of 3 x3, the output of each convolutional layer is called a feature map, the feature map is a three-dimensional array formed by the size (for example, 32x32) and the number of channels (for example, 64), and one convolutional layer can receive the feature map output by the last convolutional layer as input and output a new feature map; in the convolutional layer, a solid line indicates that a feature map at the starting point and a feature map at the end point are fused, and the fusion mode is that numerical values of corresponding positions on the feature maps are added; the dotted line indicates that the convolutional layer at the starting point reduces the size of the feature map to half of the original size, increases the number of channels of the feature map to twice of the original size (see the position of the first dotted line in fig. 5: the size of the feature map is changed from 32x32 to 16x16, and the number of channels of the feature map is changed from 64 to 128), and then fuses the feature map at the starting point with the feature map at the end point in a manner consistent with the solid line. In this embodiment, an input picture is 32x32 in size and is divided into three RGB channels, and the input picture is preprocessed by a convolutional layer to obtain feature maps of 64 channels and 32x32 in size, and then is processed by 16 convolutional layers with solid line or virtual line structures to obtain feature maps of 4x4 in size and 512 in number, the feature maps pass through an average pooling layer named "avg pool", the pooling layer calculates an average value of 16 values in 512 channels to obtain 512 values, the obtained 512 values are input into an "FC 10" layer, and the "FC 10" layer finally outputs 2 values corresponding to probability values of "normal tobacco tip" and "abnormal tobacco tip".
The fourth step: and performing iterative training on the preprocessed leaf tip abnormity classification image sample set by adopting the convolution neural network in the last step to obtain the trained tobacco leaf tip abnormity classification convolution neural network.
The specific iterative training process is as follows:
the preprocessed abnormal leaf tip classification image sample set is used for training a full convolution neural network, when the network is trained, a cross entropy function CrossEntropyLoss is adopted as a loss function, and the formula is as follows:
y in the formulai ylabelRespectively "output obtained after inputting pictures into the network model" and "label", where n is the logarithm of the pictures in the training data (one input picture and one corresponding label form a pair). In the training process, when the current iteration error of the last two times tends to be unchanged, the network confirms that the iteration error is converged, the training is stopped, and the trained tobacco leaf tip abnormity classification convolutional neural network is obtained.
In this embodiment, the iterative training process is referred to as crossEntropyLoss<10-8And when the ratio of the error values of the two times is more than 99.999%, the network is judged to be converged and the training is stopped.
And after the trained tobacco tip abnormity classification convolutional neural network is obtained, the tobacco tip image collected in the step S3 is input into the trained tobacco tip abnormity convolutional neural network to output probability values of normal tobacco tip and abnormal tobacco tip.
Step S5: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
When the probability value of the normal tip part of the tobacco leaves is larger than the probability value of the abnormal tip part of the tobacco leaves, outputting the state recognition result of the tip part of the tobacco leaves as that the tip part of the tobacco leaves is normal;
and when the probability value of the normal tip part of the tobacco leaves is less than or equal to the probability value of the abnormal tip part of the tobacco leaves, outputting the state recognition result of the tip part of the tobacco leaves as the abnormal tip part of the tobacco leaves.
The embodiment provides a method for detecting abnormality of a tobacco tip in a tobacco leaf baking process, which mainly aims at identifying the tobacco leaf state in a tobacco leaf yellowing stage, a convolutional neural network is used for positioning the position of the tobacco tip and classifying whether the tobacco tip is abnormal, and when the acquired data volume is large enough, a model has strong robustness; with the increasing data volume of the training model, the performance of the model can be stably improved;
meanwhile, according to the method for detecting the abnormality of the tobacco leaf tip in the tobacco leaf baking process, the convolution neural network is used for predicting the thermodynamic diagram, then the position of the tobacco leaf tip is located in a mode of finding a local maximum value in the thermodynamic diagram, and the accuracy is greatly improved; meanwhile, because the thermodynamic diagram is used as the label, each pixel on the diagram can provide supervision information when the network is trained, and a large amount of supervision information can enable the network to be effectively converged under a small training sample set.
In order to effectively improve the accuracy and reliability of the tobacco leaf abnormality identification and effectively improve the automation degree and efficiency in the tobacco leaf abnormality identification process, the application provides an embodiment of a detection device for the tobacco leaf tip abnormality in the tobacco leaf curing process, which is wholly or partially contained in a detection method for the tobacco leaf tip abnormality in the tobacco leaf curing process, and referring to fig. 6, the detection device for the tobacco leaf tip abnormality in the tobacco leaf curing process comprises the following contents:
a tobacco leaf image acquisition module: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera;
a tobacco leaf thermodynamic diagram acquisition module: inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram;
a tobacco tip identification module: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
the tobacco leaf tip abnormal probability output module: inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain probability values of normal tobacco leaf tip and abnormal tobacco leaf tip;
the tobacco leaf tip abnormity identification output module: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
The detection device for the abnormality of the tip part of the tobacco leaf in the tobacco leaf baking process in the implementation realizes automatic identification, is low in cost, simple, convenient and obvious in effect, can be applied to the tobacco leaf baking control process, is high in identification precision, and improves the quality of the tobacco leaf.
In order to effectively improve the accuracy and reliability of tobacco leaf abnormity identification and effectively improve the automation degree and efficiency of the tobacco leaf abnormity identification process, the application provides a terminal of all or part of contents in the method for detecting the tobacco leaf tip abnormity in the tobacco leaf baking process, and the terminal specifically comprises the following contents:
the processor executes the program to realize the steps of the method for detecting the abnormal tip part of the tobacco leaves in the tobacco leaf curing process.
A communication interface and a bus; the processor and the memory complete mutual communication through a communication interface and a bus; the terminal can be a desktop computer, a tablet computer, a mobile terminal and the like.
In a specific implementation, the present application further provides a computer storage medium, where the computer storage medium may store a program, and the computer program, when executed by a processor, may implement some or all of the steps of the method for detecting the abnormality of the tip portion of the tobacco during the tobacco flue-curing process provided by the present application. The computer storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM) or a Random Access Memory (RAM).
According to the tobacco leaf abnormity identification method, the target detection algorithm based on the convolutional neural network is adopted mainly aiming at the tobacco leaf state identification in the tobacco leaf yellowing stage, meanwhile, Gaussian interpolation calculation is utilized, high-precision automatic tobacco leaf tip identification and extraction are achieved, and the accuracy of tobacco leaf detection is improved.
Those skilled in the art will clearly understand that the techniques in the embodiments of the present application may be implemented by way of software plus a required general hardware platform. Based on such understanding, the technical solutions in the embodiments of the present application may be essentially implemented or a part contributing to the prior art may be embodied in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the embodiments or some parts of the embodiments of the present application.
The same and similar parts in the various embodiments in this specification may be referred to each other. In particular, as for the apparatus embodiment, since it is substantially similar to the method embodiment, the description is simple, and the relevant points can be referred to the description in the method embodiment.
Although the present invention has been described in detail with reference to the foregoing examples, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention as defined in the following claims. All modifications, equivalents and the like which come within the spirit and principle of the invention are intended to be included within the scope of the invention.
Claims (6)
1. A method for detecting abnormality of a tobacco leaf tip in a tobacco leaf curing process is characterized by comprising the following steps:
s1: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera;
s2: inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram;
the trained tobacco leaf tip positioning convolution neural network construction steps are as follows:
the first step is as follows: acquiring a blade tip positioning image sample set through a camera;
the second step is that: preprocessing the blade tip positioning image sample set, interpolating the position with the cigarette tip by taking a Gaussian function as a template, and manufacturing a blade tip positioning image sample set label;
the manufacturing steps of the label are as follows:
firstly, creating a blank image set which is in one-to-one correspondence with images in the blade tip positioning image sample set, wherein the size of the images in the blank image set is consistent with that of the images in the blade tip positioning image sample set;
secondly, marking the position of the tip of the tobacco leaf in each image in the blade tip positioning image sample set;
thirdly, the label in each image is manufactured in a mode that the position corresponding to the position of the tip of the tobacco leaf on the corresponding blank image is determined according to the position of the tip of the tobacco leaf, interpolation is carried out according to the position corresponding to the position of the tip of the tobacco leaf in the blank image by taking a two-dimensional Gaussian function as a template, the central value of the two-dimensional Gaussian function is the position of the corresponding tip of the tobacco leaf in the blank image, and when at least 2 two-dimensional Gaussian functions have overlapping areas, the maximum value on the corresponding Gaussian function is taken as the position of the corresponding tip of the tobacco leaf in the blank image in the overlapping areas;
finally, making a corresponding label for each image in the blank image set;
the third step: constructing a tobacco leaf tip positioning convolution neural network;
the fourth step: performing iterative training on the preprocessed blade tip positioning image sample set by adopting the convolutional neural network in the previous step to obtain the trained tobacco tip positioning convolutional neural network;
s3: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
in the tobacco leaf thermodynamic diagram, the value of each pixel point represents the probability that the pixel point has a tobacco tip part, when the value of the pixel point is more than or equal to the value of the pixel point in a 15 x 15 area taking the pixel point as the center, the value of the pixel point is the local maximum value, the pixel point corresponding to the local maximum value is the tobacco leaf tip part position, the tobacco leaf tip part position in the thermodynamic diagram is taken as the center, and a tobacco leaf tip part picture is intercepted in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
s4: inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain probability values of normal tobacco leaf tip and abnormal tobacco leaf tip;
s5: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
2. The method for detecting the abnormality of the tobacco leaf tip in the tobacco leaf curing process according to claim 1, wherein the tobacco leaf tip positioning convolutional neural network is a stacked hourglass network formed by 2 densely connected hourglass structure networks, and the output is a tobacco leaf thermodynamic diagram.
3. The method for detecting the abnormality of the tip portion of the tobacco leaf in the tobacco leaf curing process according to claim 1, wherein the step of constructing the classified convolutional neural network of the abnormality of the tip portion of the tobacco leaf in the step S4 comprises the steps of:
the first step is as follows: acquiring a tobacco leaf image sample set through a camera, and marking and intercepting a leaf tip part area in the tobacco leaf image set to obtain a leaf tip abnormal classification image sample set;
the second step is that: preprocessing the abnormal leaf tip classified image sample set, manufacturing an abnormal leaf tip classified image sample set label, classifying the abnormal leaf tip classified image sample set according to whether the tip of the tobacco leaves in the abnormal leaf tip classified image sample set is blackened, wherein the label is that the tip of the tobacco leaves is abnormal, and otherwise, the tip of the tobacco leaves is normal;
the third step: constructing a tobacco leaf abnormal classification convolutional neural network;
the fourth step: and performing iterative training on the preprocessed leaf tip abnormity classification image sample set by adopting the convolution neural network in the last step to obtain the trained tobacco leaf tip abnormity classification convolution neural network.
4. The method according to claim 3, wherein the tobacco leaf abnormality classification convolutional neural network is formed by sequentially connecting 1 average pooling layer of 16 convolutional layers with convolutional kernel size of 3 x3 and 1 full-connection layer, and the full-connection layer outputs 2 values, which correspond to the probability values of normal tobacco leaf tip and abnormal tobacco leaf tip.
5. The method according to claim 1, wherein in step S5, when the probability value of the normal tip of the tobacco leaf is greater than the probability value of the abnormal tip of the tobacco leaf, the result of identifying the state of the tip of the tobacco leaf is output as the normal tip of the tobacco leaf; and when the probability value of the normal tip part of the tobacco leaves is less than or equal to the probability value of the abnormal tip part of the tobacco leaves, outputting the state recognition result of the tip part of the tobacco leaves as the abnormal tip part of the tobacco leaves.
6. A detection device for the abnormality of the tip portion of a tobacco leaf in the tobacco leaf curing process comprises:
a tobacco leaf image acquisition module: acquiring a tobacco leaf image in the tobacco leaf baking process through a camera;
a tobacco leaf thermodynamic diagram acquisition module: inputting the tobacco leaf image into a trained tobacco leaf tip positioning convolution neural network to obtain a tobacco leaf thermodynamic diagram;
the trained tobacco leaf tip positioning convolution neural network construction steps are as follows:
the first step is as follows: acquiring a blade tip positioning image sample set through a camera;
the second step is that: preprocessing the blade tip positioning image sample set, interpolating the position with the cigarette tip by taking a Gaussian function as a template, and manufacturing a blade tip positioning image sample set label;
the manufacturing steps of the label are as follows:
firstly, creating a blank image set which is in one-to-one correspondence with images in the blade tip positioning image sample set, wherein the size of the images in the blank image set is consistent with that of the images in the blade tip positioning image sample set;
secondly, marking the position of the tip of the tobacco leaf in each image in the blade tip positioning image sample set;
thirdly, the label in each image is manufactured in a mode that the position corresponding to the position of the tip of the tobacco leaf on the corresponding blank image is determined according to the position of the tip of the tobacco leaf, interpolation is carried out according to the position corresponding to the position of the tip of the tobacco leaf in the blank image by taking a two-dimensional Gaussian function as a template, the central value of the two-dimensional Gaussian function is the position of the corresponding tip of the tobacco leaf in the blank image, and when at least 2 two-dimensional Gaussian functions have overlapping areas, the maximum value on the corresponding Gaussian function is taken as the position of the corresponding tip of the tobacco leaf in the blank image in the overlapping areas;
finally, making a corresponding label for each image in the blank image set;
the third step: constructing a tobacco leaf tip positioning convolution neural network;
the fourth step: performing iterative training on the preprocessed blade tip positioning image sample set by adopting the convolutional neural network in the previous step to obtain the trained tobacco tip positioning convolutional neural network;
a tobacco tip identification module: determining a local maximum value on the tobacco leaf thermodynamic diagram to obtain a tobacco leaf tip part position, and intercepting a tobacco leaf tip part picture at the tobacco leaf tip part position in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
in the tobacco leaf thermodynamic diagram, the value of each pixel point represents the probability that the pixel point has a tobacco tip part, when the value of the pixel point is more than or equal to the value of the pixel point in a 15 x 15 area taking the pixel point as the center, the value of the pixel point is the local maximum value, the pixel point corresponding to the local maximum value is the tobacco leaf tip part position, the tobacco leaf tip part position in the thermodynamic diagram is taken as the center, and a tobacco leaf tip part picture is intercepted in a tobacco leaf image corresponding to the tobacco leaf thermodynamic diagram;
the tobacco leaf tip abnormal probability output module: inputting the tobacco leaf tip image into a trained tobacco leaf tip abnormity classification convolution neural network to obtain probability values of normal tobacco leaf tip and abnormal tobacco leaf tip;
the tobacco leaf tip abnormity identification output module: and outputting a recognition result of the tip part state of the tobacco leaves, wherein the recognition result comprises normal tip part of the tobacco leaves and abnormal tip part of the tobacco leaves.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110205229.9A CN112926648B (en) | 2021-02-24 | 2021-02-24 | Method and device for detecting abnormality of tobacco leaf tip in tobacco leaf baking process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110205229.9A CN112926648B (en) | 2021-02-24 | 2021-02-24 | Method and device for detecting abnormality of tobacco leaf tip in tobacco leaf baking process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112926648A CN112926648A (en) | 2021-06-08 |
| CN112926648B true CN112926648B (en) | 2021-11-16 |
Family
ID=76170663
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110205229.9A Active CN112926648B (en) | 2021-02-24 | 2021-02-24 | Method and device for detecting abnormality of tobacco leaf tip in tobacco leaf baking process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112926648B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113468975A (en) * | 2021-06-09 | 2021-10-01 | 浙江大华技术股份有限公司 | Fighting behavior detection method and device |
| CN113808084A (en) * | 2021-08-25 | 2021-12-17 | 杭州安脉盛智能技术有限公司 | Model-fused online tobacco bale surface mildew detection method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109540894A (en) * | 2018-12-17 | 2019-03-29 | 云南省烟草公司红河州公司 | A kind of lossless rapid detection method of cured tobacco leaf maturity |
| CN111191622A (en) * | 2020-01-03 | 2020-05-22 | 华南师范大学 | Posture recognition method and system based on thermodynamic diagram and offset vector and storage medium |
| CN112163527A (en) * | 2020-09-29 | 2021-01-01 | 华中科技大学 | Fusion model-based tobacco leaf baking state identification method, device and system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA3061912A1 (en) * | 2017-05-08 | 2018-11-15 | PlantSnap, Inc. | Systems and methods for electronically identifying plant species |
| CN110082399A (en) * | 2019-04-25 | 2019-08-02 | 云南中烟工业有限责任公司 | A kind of fast non-destructive detection method of fresh tobacco leaves total reducing sugar |
-
2021
- 2021-02-24 CN CN202110205229.9A patent/CN112926648B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109540894A (en) * | 2018-12-17 | 2019-03-29 | 云南省烟草公司红河州公司 | A kind of lossless rapid detection method of cured tobacco leaf maturity |
| CN111191622A (en) * | 2020-01-03 | 2020-05-22 | 华南师范大学 | Posture recognition method and system based on thermodynamic diagram and offset vector and storage medium |
| CN112163527A (en) * | 2020-09-29 | 2021-01-01 | 华中科技大学 | Fusion model-based tobacco leaf baking state identification method, device and system |
Non-Patent Citations (1)
| Title |
|---|
| 基于光谱和形状特征的水稻扫描叶片氮素营养诊断;顾清 等;《农业机械学报》;20120831;第43卷(第8期);第170-174页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112926648A (en) | 2021-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| RU2680765C1 (en) | Automated determination and cutting of non-singular contour of a picture on an image | |
| CN112926648B (en) | Method and device for detecting abnormality of tobacco leaf tip in tobacco leaf baking process | |
| CN111950453A (en) | Optional-shape text recognition method based on selective attention mechanism | |
| CN111369605B (en) | Infrared and visible light image registration method and system based on edge features | |
| CN111612817A (en) | Target tracking method based on depth feature adaptive fusion and context information | |
| CN112949704B (en) | Tobacco leaf maturity state identification method and device based on image analysis | |
| CN112884782B (en) | Biological object segmentation method, apparatus, computer device, and storage medium | |
| CN112581462A (en) | Method and device for detecting appearance defects of industrial products and storage medium | |
| CN110310305B (en) | Target tracking method and device based on BSSD detection and Kalman filtering | |
| CN111768415A (en) | Image instance segmentation method without quantization pooling | |
| CN114140623A (en) | Image feature point extraction method and system | |
| CN116385707A (en) | Deep learning scene recognition method based on multi-scale features and feature enhancement | |
| CN115410030A (en) | Target detection method, target detection device, computer equipment and storage medium | |
| CN116664892A (en) | Multi-temporal remote sensing image registration method based on cross attention and deformable convolution | |
| CN115239672A (en) | Defect detection method and device, equipment and storage medium | |
| CN113421210B (en) | Surface point Yun Chong construction method based on binocular stereoscopic vision | |
| CN117036756A (en) | Remote sensing image matching method and system based on variation automatic encoder | |
| CN109800815B (en) | Training method, wheat recognition method and training system based on random forest model | |
| CN116469095A (en) | Method for detecting self-adaptive three-dimensional target of space-time scene by using radar sensing fusion | |
| CN116129417A (en) | Digital instrument reading detection method based on low-quality image | |
| CN113051901B (en) | Identification card text recognition method, system, medium and electronic terminal | |
| Hernandez et al. | Classification of color textures with random field models and neural networks | |
| CN114926826A (en) | Scene text detection system | |
| CN114155575A (en) | Method, device, medium and equipment for identifying certificate photo lipstick based on convolutional neural network | |
| CN113793372A (en) | Optimal registration method and system for different-source images |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |