A method for identifying and removing cigarette debris based on images collected by area array cameras
Through the surface array camera, the image acquisition and combined with light adjustment and deep learning models, the problem of difficult debris identification in the existing technology is solved, the recognition accuracy and efficiency are improved, and the tobacco processing industry is used to adapt to the purity of tobacco sheets. High demand.
Patent Information
- Application Number
- CN202411225845.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-09-03
AI Technical Summary
The prior art has problems with low recognition accuracy and efficiency when identifying and removing small volumes and small areas of debris, especially in the tobacco processing industry where the purity of tobacco sheets is increasingly stringent.
The smoke debris recognition and removal method based on the surface array camera collects images, and light adjustment is performed by obtaining lighting equipment information, setting a reference model and lighting simulation model, combining color abnormality recognition and vein image processing, debris features are extracted for recognition, and Debris recognition and removal are performed through deep learning models.
It significantly improves the accuracy of identification of small volumes and small areas of debris, reduces interference factors, adapts to smoke sheet detection in various situations, and ensures the quality of collected smoke sheet images.
Smart Images

Figure CN118762355B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of cigarette debris recognition and removal, and in particular relates to a cigarette debris recognition and removal method based on images collected by an area array camera. Background Art
[0002] In the tobacco processing industry, ensuring the purity of tobacco flakes (or stems) is the key to improving product quality and consumer satisfaction. Traditionally, large and obvious debris, such as large pieces of metal and plastic, can be effectively identified and removed through simple physical screening or manual selection. However, with the improvement of production automation and consumers' increasingly stringent requirements for product quality, the identification and removal of small and small debris has become a problem that needs to be solved urgently.
[0003] In the prior art, there are already a variety of debris recognition and removal systems based on image processing. Although they can recognize and remove large and obvious debris, they often face many challenges when dealing with small and small debris. On the one hand, due to the small size and area of the debris, its features in the image may be relatively weak; on the other hand, the image processing algorithms and recognition models of the existing systems may not fully consider the characteristics of small debris, resulting in low recognition accuracy and removal efficiency.
[0004] Based on this, in order to realize the recognition and removal of small debris, the present invention provides a method for recognizing and removing cigarette debris based on images collected by an area array camera. Summary of the invention
[0005] In order to solve the problems existing in the above-mentioned scheme, the present invention provides a method for identifying and removing cigarette debris based on images collected by an area array camera.
[0006] The purpose of the present invention can be achieved through the following technical solutions:
[0007] A method for identifying and removing cigarette debris based on images collected by an area array camera, the method comprising:
[0008] Step 1: Obtain lighting equipment information, and adjust lighting according to the lighting equipment information;
[0009] Furthermore, the method for adjusting the lighting includes:
[0010] Set the reference model and the reference standard image corresponding to the reference model;
[0011] During the smoke sheet detection process, an image of the reference model is collected in real time and marked as a reference collected image;
[0012] Compare the reference captured image with the reference standard image to determine whether the illumination adjustment meets the image capture requirements;
[0013] When it is judged that the image acquisition requirements are met, no corresponding operation is performed;
[0014] When it is determined that the image acquisition requirements are not met, corresponding lighting adjustments are made.
[0015] Furthermore, the setting method of the reference model includes:
[0016] Obtaining each sundry object image and each cigarette piece image, and determining each image feature according to each sundry object image and each cigarette piece image;
[0017] Establish a lighting simulation model, the expression of the lighting simulation model is:
[0018] ;
[0019] Where: x is the input data, the input data is the two image features to be compared; the output data is the illumination simulation value GL(x);
[0020] The pairwise image features are analyzed through the illumination simulation model to obtain the illumination simulation value between the pairwise image features;
[0021] The image features whose illumination simulation values are all 1 are classified into one category, which is marked as the feature simulation category;
[0022] An image feature is selected from each feature simulation class as an image representative feature, and a reference model is set according to each image representative feature.
[0023] Step 2: Capture the image of the cigarette sheet using the area array camera to obtain the image of the cigarette sheet;
[0024] Step 3: performing color anomaly recognition on the cigarette sheet image to obtain a color anomaly recognition result, and determining the foreign matter according to the color anomaly recognition result;
[0025] Furthermore, the method for identifying color anomalies in a cigarette sheet image includes:
[0026] Determine the various tobacco colors of the tobacco according to the historically collected images of the tobacco, and establish a tobacco reserve according to the various tobacco colors;
[0027] A color anomaly recognition model is established based on the cigarette sheet reserve. The expression of the color anomaly recognition model is:
[0028] ;
[0029] Where: s is the input data, the input data is the color of the cigarette; the output data is the color outlier value YW(s);
[0030] The color anomaly recognition model is used to identify and analyze each cigarette sheet image, and the color anomaly value corresponding to each cigarette sheet image is obtained;
[0031] When the color outlier value is 0, the color recognition is evaluated to be normal;
[0032] When the color outlier value is 1, the color recognition anomaly is evaluated.
[0033] Step 4: grayscale processing is performed on the cigarette sheet image to obtain a cigarette sheet grayscale image, and a vein image is determined according to the cigarette sheet grayscale image;
[0034] Furthermore, the method for determining the vein image according to the grayscale image of the cigarette sheet includes:
[0035] Step SA1: Identify the starting point of veins in the grayscale image of the cigarette sheet;
[0036] Step SA2: Identify the grayscale value of the pixel corresponding to the starting point of the vein and mark it as the starting grayscale value; mark the pixels adjacent to the vein as adjacent pixels, and mark the grayscale values of the adjacent pixels as adjacent grayscale values;
[0037] Marking the vein starting point and each neighboring pixel in the cigarette grayscale image; evaluating each neighboring pixel in the cigarette grayscale image through a preset vein judgment model to obtain a vein judgment value corresponding to each neighboring pixel;
[0038] The neighboring pixels with a context judgment value of 1 are marked as context pixels;
[0039] Mark the neighboring pixels with a context judgment value of 0 as additional pixels;
[0040] Mark the vein pixels and the extra pixels in the smoke sheet grayscale image accordingly;
[0041] Step SA3: determine new neighboring pixels, mark each neighboring pixel in the cigarette grayscale image, analyze it through the vein judgment model, and obtain the vein judgment value corresponding to each neighboring pixel;
[0042] The neighboring pixels with a context judgment value of 1 are marked as context pixels;
[0043] Mark the neighboring pixels with a context judgment value of 0 as additional pixels;
[0044] Mark the vein pixels and the extra pixels in the smoke sheet grayscale image accordingly;
[0045] Step SA4: loop step SA3 until there are no more vein pixels;
[0046] Step SA5: determining whether there is a new vein starting point according to the grayscale image of the cigarette sheet;
[0047] When it is determined that there is a new context starting point, return to step SA2;
[0048] When it is determined that there is no new vein starting point, the vein pixels in the tobacco leaf grayscale image are connected and merged to form leaf veins, and the current tobacco leaf grayscale image is marked as a vein image.
[0049] Furthermore, the expression of the context judgment model is:
[0050] ;
[0051] Where: q is the input data; the output data is the context judgment value MW(q).
[0052] Step 5: Identify debris based on the vein image to obtain the debris identification result;
[0053] Furthermore, the method for identifying debris based on the vein image includes:
[0054] Identify the vein defect area according to the vein image, extract the regional defect features corresponding to the vein defect area, determine whether it is debris according to the regional defect features, and obtain the debris identification result corresponding to the leaf vein area;
[0055] Based on the leaf veins in the vein image, a simulation evaluation is performed on the non-vein area to determine whether there are debris and obtain the debris recognition result corresponding to the non-vein area.
[0056] Step 6: Remove the identified debris.
[0057] Compared with the prior art, the present invention has the following beneficial effects:
[0058] The present invention optimizes the debris recognition method and makes a special design based on the characteristics of small-volume and small-area debris, which effectively solves the problem of difficulty in identifying such debris in the prior art. It can capture and identify these tiny debris more accurately and significantly improve the recognition accuracy. By setting a reference model for lighting adjustment, a large number of interference factors will be reduced, and it can adapt to cigarette sheet detection in various situations. The lighting can be adjusted directly based on the results to ensure that the collected cigarette sheet images meet the image acquisition requirements. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0060] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0061] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0062] like Figure 1 As shown, a method for identifying and removing cigarette debris based on images collected by an area array camera includes:
[0063] Step 1: Obtain lighting equipment information and adjust lighting according to the lighting equipment information;
[0064] Lighting equipment information includes equipment control method, type, model and other related information;
[0065] In one embodiment, the lighting is adjusted according to the environmental information;
[0066] Because the imaging quality of the area array camera is significantly affected by lighting conditions. In the case of insufficient or excessive lighting, the image may have problems such as loss of dark details and overexposure of bright parts, which will affect the recognition accuracy of debris. Therefore, it is necessary to adjust the lighting according to environmental information; obtain environmental information, which refers to the environmental information such as lighting and temperature corresponding to the smoke sheet transmission detection, and adjust the lighting according to the environmental information.
[0067] Lighting adjustment according to environmental information can be performed based on existing technologies, that is, using existing methods to adjust lighting to ensure that the lighting meets the collection conditions.
[0068] In one embodiment, the illumination adjustment may be performed in the following manner, including:
[0069] A reference model is set according to the actual situation of smoke sheet detection, and the reference model is a physical model set in the smoke sheet detection area; an image of the reference model is collected in real time during the smoke sheet detection process and marked as a reference collected image; a reference standard image of the reference model is set; the reference collected image is compared with the reference standard image, and it is determined whether the illumination adjustment meets the image acquisition requirements according to the difference between the two images;
[0070] When it is judged that the image acquisition requirements are met, no corresponding operation is performed;
[0071] When it is judged that the image acquisition requirements are not met, corresponding lighting adjustments are made; when it is determined that the image acquisition requirements are not met, how to make adjustments is determined based on the data of the pre-illumination simulation; that is, the image of the reference model under different lighting conditions is first simulated, and then it is determined how to adjust to meet the acquisition conditions.
[0072] By setting a reference model for lighting adjustment, a large number of interference factors will be reduced, adapting to cigarette sheet detection in various situations, and directly adjusting the lighting based on the results to ensure that the collected cigarette sheet images meet the image acquisition requirements.
[0073] The reference model setting methods include:
[0074] Obtain possible debris images and cigarette flake images, and determine the features of each image, such as color, distribution combination of each color, etc., based on each debris image and cigarette flake image;
[0075] A lighting simulation model is established. The lighting simulation model is used to analyze the pairwise image features to determine whether the effects of lighting on the pairwise image features are the same as those on the acquisition, that is, under the current acquisition conditions, whether the pairwise image features are affected by lighting in the same way during image acquisition; the same lighting effects are considered to meet the lighting simulation requirements; specifically, the lighting simulation requirements can be judged based on existing intelligent technologies, field simulations, historical data analysis, etc.; the expression of the lighting simulation model is: ; Where: x is the input data, the input data is the two image features to be compared; the output data is the illumination simulation value GL(x);
[0076] The pairwise image features are analyzed through the illumination simulation model to obtain the illumination simulation value between the pairwise image features;
[0077] The image features whose illumination simulation values are all 1 are classified into one category, which is marked as the feature simulation category, that is, the illumination simulation values between the image features in the feature simulation category are all 1;
[0078] Select an image feature from each feature simulation class as the image representative feature. You can choose any one or select an image feature that is obvious as the image representative feature. Set a reference model based on each image representative feature; that is, each image representative feature is reflected in the reference model; such as a plane block, set each image representative feature on the plane block, and mark the plane block as the reference model.
[0079] Step 2: Capture the image of the cigarette sheet using the area array camera to obtain the image of the cigarette sheet;
[0080] Step 3: Perform color anomaly recognition on the cigarette image to obtain a color anomaly recognition result, and determine the debris based on the color anomaly recognition result;
[0081] The method for identifying color anomalies in cigarette images includes:
[0082] Determine the possible colors of the tobacco sheet according to the historical tobacco sheet collection images, including the color of the pathological changes in a certain part of the tobacco sheet, etc., mark them as tobacco sheet colors, and establish a tobacco sheet reserve according to each tobacco sheet color, that is, the tobacco sheet reserve is used to store each tobacco sheet color;
[0083] A color anomaly recognition model is established based on the cigarette sheet reserve. The color anomaly recognition model is used to compare the cigarette sheet image with the colors of each cigarette sheet in the cigarette sheet reserve to determine whether there is a color that does not belong to the cigarette sheet color. If so, it is considered a color anomaly. The expression of the color anomaly recognition model is: ; Where: s is the input data, the input data is the color of the cigarette; the output data is the color outlier value YW(s);
[0084] The color anomaly recognition model is used to identify and analyze each cigarette sheet image, and the color anomaly value corresponding to each cigarette sheet image is obtained;
[0085] When the color outlier value is 0, the color recognition is evaluated to be normal;
[0086] When the color outlier value is 1, the color recognition is evaluated as abnormal. At this time, the debris can be determined based on the color abnormal area.
[0087] Step 4: grayscale processing is performed on the cigarette sheet image to obtain a cigarette sheet grayscale image, and the cigarette sheet grayscale image is analyzed to determine the vein image corresponding to the cigarette sheet;
[0088] The method for analyzing the grayscale image of cigarette pieces includes:
[0089] Step SA1: Identify the vein starting point in the grayscale image of the tobacco leaf. The vein starting point is the position at one end of the tobacco leaf where the vein diameter is the largest. It can be directly determined based on the tobacco leaf image. The vein starting point is also a vein pixel.
[0090] Step SA2: Identify the gray value of the pixel corresponding to the starting point of the vein and mark it as the starting point gray value;
[0091] Identify the grayscale value corresponding to each pixel in the grayscale image of the cigarette sheet, mark the pixels adjacent to the vein as adjacent pixels, that is, the pixels adjacent to the starting point of the vein are also adjacent pixels; mark the grayscale value of the adjacent pixel as the adjacent grayscale value;
[0092] Identify the neighboring grayscale values corresponding to the grayscale value of the starting point, analyze the neighboring grayscale values through the preset vein judgment model, and judge whether the neighboring pixels belong to the vein of the cigarette sheet, that is, judge according to the distribution and value change trend of the neighboring grayscale values, the grayscale values of the corresponding veins and the grayscale values of other adjacent pixels. Because the grayscale of the veins is different on the cigarette sheet, it can be judged according to its grayscale change. The expression of the vein judgment model is: , where: q is the input data; the output data is the context judgment value;
[0093] Mark the vein starting point and each neighboring pixel in the grayscale image of the cigarette sheet; evaluate each neighboring pixel through the vein judgment model to obtain the vein judgment value corresponding to each neighboring pixel;
[0094] The neighboring pixels with a context judgment value of 1 are marked as context pixels;
[0095] Mark the neighboring pixels with a context judgment value of 0 as additional pixels;
[0096] Mark the vein pixels and the extra pixels in the smoke sheet grayscale image accordingly;
[0097] Step SA3: determine new neighboring pixels, excluding vein pixels and extra pixels, based on vein pixels, mark each neighboring pixel in the cigarette grayscale image, analyze through the vein judgment model, and obtain the vein judgment value corresponding to each neighboring pixel;
[0098] The neighboring pixels with a context judgment value of 1 are marked as context pixels;
[0099] Mark the neighboring pixels with a context judgment value of 0 as additional pixels;
[0100] Mark the vein pixels and the extra pixels in the smoke sheet grayscale image accordingly;
[0101] Step SA4: loop step SA3 until there are no more vein pixels;
[0102] Step SA5: Determine whether there is a new vein starting point according to the grayscale image of the cigarette sheet. This is mainly for the case where the vein is interrupted due to debris, holes, etc. Since there are veins later, the vein pixels can be determined again. Specifically, all pixels along the vein pixel direction are regarded as adjacent pixels to determine whether they are vein pixels. If so, mark the first one as the vein starting point.
[0103] When it is determined that there is a new context starting point, return to step SA2;
[0104] When it is determined that there is no new vein starting point, the vein pixels in the tobacco leaf grayscale image are connected and merged to form leaf veins, and the current tobacco leaf grayscale image is marked as a vein image.
[0105] Step 5: Identify debris based on the vein image to obtain the debris identification result;
[0106] Methods for identifying debris based on vein images include:
[0107] Identify the vein defect area according to the vein image, extract the regional defect features corresponding to the vein defect area, including gray value distribution, shape and other features, and judge whether it is debris according to the regional defect features; that is, match and judge according to the features corresponding to the possible debris and non-debris features such as holes, generally, the non-debris features that the tobacco leaves may have, such as holes, are screened, and then determine whether it is debris, and match it through the preset corresponding feature library; obtain the debris recognition result corresponding to the leaf vein area;
[0108] Based on the leaf veins in the vein image, a simulation judgment is made to determine whether there are debris, and the debris recognition result corresponding to the non-vein area is obtained.
[0109] The simulation judgment is based on the leaf veins in the vein image, that is, the pixel values of the leaf veins and the additional pixels adjacent to them are simulated, and the grayscale simulation of the tobacco leaves in the non-vein areas is performed according to the actual possible changes of the tobacco leaves to determine the corresponding grayscale distribution, and then judge whether it is debris. For larger areas of damaged and deteriorated areas, they will also be regarded as debris according to the grayscale changes of normal tobacco leaves. Therefore, the corresponding areas can be removed simultaneously to ensure the quality of the tobacco leaves.
[0110] Exemplarily, a sample data set is created, the sample data set includes a material sample and a manually annotated sample set, the material sample is a vein image set processed based on a cigarette sheet image, the manually annotated sample set is a picture set after the vein image set is format converted and manually annotated with debris, the pictures in the manually annotated sample set are binarized and saved in a single-channel form, the material sample set and the manually annotated sample set are divided according to proportion to form a second photo sample set and a second manually annotated sample set; a LinkNet network model is built based on the LinkNet network structure under the Pytorch deep learning framework, the parameters of the LinkNet network model are set, the second photo sample set and the second manually annotated sample set are input into the LinkNet network model, and the LinkNet network model is trained based on the Pytorch deep learning framework, multiple models are saved during the training process, and the model with the smallest error is selected as the cigarette sheet simulation model using the validation set data; simulation analysis is performed through the cigarette sheet simulation model to obtain the debris recognition result.
[0111] Specifically, LinkNet adopts the idea of autoencoder, and its architecture is divided into two parts: encoder and decoder. The encoder encodes the input into a low-dimensional space, and the decoder reconstructs the input from the low-dimensional space. It consists of an initial block, a final block, an encoder block with four convolutional modules, and a decoder block with four deconvolutional modules.
[0112] 1. Image preprocessing
[0113] Before inputting into LinkNet, the image is preprocessed to ensure that the input image meets the requirements and expectations of the model (the input image is 3*192*1408);
[0114] Resizing: Resize the input image to a standard size that the network can handle.
[0115] Normalization: Normalize the input image and scale the pixel values to a certain range (0-1). This helps accelerate model convergence and improve training stability.
[0116] 2. Encoder
[0117] Feature extraction: The input preprocessed image is passed through the pre-trained convolutional neural network ResNet18 as an encoder to extract the feature information of the image. The encoder consists of 13 3×3 convolutional layers and 5 2×2 pooling layers. Batch normalization is used between each convolutional layer, followed by ReLU nonlinearity, which gradually reduces the size of the feature map and increases the level of abstraction of the feature.
[0118] 3. Decoder
[0119] Upsampling: In the decoder, an upsampling operation is used to enlarge the low-resolution feature map extracted by the encoder to the size of the original input image. This is achieved through a transposed convolution operation.
[0120] Skip Connections: One of the features of LinkNet is the use of skip connection technology, which connects the feature map in the encoder with the feature map of the corresponding level in the decoder to preserve and utilize multi-level feature information.
[0121] Feature fusion: The encoder features obtained by skip connections are fused with the features of the decoder to enrich the semantic information of the decoder.
[0122] 4. Classification Head
[0123] Pixel classification: Finally, the feature map output by the decoder is passed through a classification head, usually a convolutional layer or a fully connected layer, to assign each pixel to a predefined semantic category. This step is the key to the semantic segmentation task and determines the final classification result of each pixel.
[0124] 5. Inference phase
[0125] Inference: In the inference phase, a new, unseen image is input and the pixel-level semantic segmentation result is obtained after forward propagation. This process can segment and classify the image in real time.
[0126] Step 6: Remove the identified debris.
[0127] The above formulas are all calculated by removing dimensions and taking numerical values. The formula is a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data.
[0128] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.
Claims
1. A method for identifying and removing cigarette debris based on images collected by an area array camera, characterized in that: Methods include: Step 1: Obtain lighting equipment information, and adjust lighting according to the lighting equipment information; Step 2: Capture the image of the cigarette sheet using the area array camera to obtain the image of the cigarette sheet; Step 3: performing color anomaly recognition on the cigarette sheet image to obtain a color anomaly recognition result, and determining the foreign matter according to the color anomaly recognition result; Step 4: grayscale processing is performed on the cigarette sheet image to obtain a cigarette sheet grayscale image, and a vein image is determined according to the cigarette sheet grayscale image; Step 5: Identify debris based on the vein image to obtain the debris identification result; Step 6: Remove the identified debris; The method for identifying color anomalies in cigarette images includes: Determine the various tobacco colors of the tobacco according to the historically collected images of the tobacco, and establish a tobacco reserve according to the various tobacco colors; A color anomaly recognition model is established based on the cigarette sheet reserve. The expression of the color anomaly recognition model is: ; Where: s is the input data, the input data is the color of the cigarette; the output data is the color outlier value YW(s); The color anomaly recognition model is used to identify and analyze each cigarette sheet image, and the color anomaly value corresponding to each cigarette sheet image is obtained; When the color outlier value is 0, the color recognition is evaluated to be normal; When the color outlier value is 1, the color recognition anomaly is evaluated; Methods for adjusting lighting include: Set the reference model and the reference standard image corresponding to the reference model; During the smoke sheet detection process, an image of the reference model is collected in real time and marked as a reference collected image; Compare the reference captured image with the reference standard image to determine whether the illumination adjustment meets the image capture requirements; When it is judged that the image acquisition requirements are met, no corresponding operation is performed; When it is judged that the image acquisition requirements are not met, corresponding lighting adjustments are made; The reference model setting methods include: Obtaining each sundry object image and each cigarette piece image, and determining each image feature according to each sundry object image and each cigarette piece image; Establish a lighting simulation model, the expression of the lighting simulation model is: ; Where: x is the input data, the input data is the two image features to be compared; the output data is the illumination simulation value GL(x); The pairwise image features are analyzed through the illumination simulation model to obtain the illumination simulation value between the pairwise image features; The image features whose illumination simulation values are all 1 are classified into one category, which is marked as the feature simulation category; An image feature is selected from each feature simulation class as an image representative feature, and a reference model is set according to each image representative feature.
2. The method for identifying and removing cigarette debris based on images collected by an area array camera according to claim 1 is characterized in that: The method for determining the vein image according to the grayscale image of the cigarette sheet includes: Step SA1: Identify the starting point of veins in the grayscale image of the cigarette sheet; Step SA2: Identify the grayscale value of the pixel corresponding to the starting point of the vein and mark it as the starting grayscale value; mark the pixels adjacent to the vein as adjacent pixels, and mark the grayscale values of the adjacent pixels as adjacent grayscale values; Marking the vein starting point and each neighboring pixel in the cigarette grayscale image; evaluating each neighboring pixel in the cigarette grayscale image through a preset vein judgment model to obtain a vein judgment value corresponding to each neighboring pixel; The neighboring pixels with a context judgment value of 1 are marked as context pixels; Mark the neighboring pixels with a context judgment value of 0 as additional pixels; Mark the vein pixels and the extra pixels in the smoke sheet grayscale image accordingly; Step SA3: determine new neighboring pixels, mark each neighboring pixel in the cigarette grayscale image, analyze it through the vein judgment model, and obtain the vein judgment value corresponding to each neighboring pixel; The neighboring pixels with a context judgment value of 1 are marked as context pixels; Mark the neighboring pixels with a context judgment value of 0 as additional pixels; Mark the vein pixels and the extra pixels in the smoke sheet grayscale image accordingly; Step SA4: loop step SA3 until there are no more vein pixels; Step SA5: determining whether there is a new vein starting point according to the grayscale image of the cigarette sheet; When it is determined that there is a new context starting point, return to step SA2; When it is determined that there is no new vein starting point, the vein pixels in the tobacco leaf grayscale image are connected and merged to form leaf veins, and the current tobacco leaf grayscale image is marked as a vein image.
3. The method for identifying and removing cigarette debris based on images collected by an area array camera according to claim 2 is characterized in that: The expression of the context judgment model is: ; Where: q is the input data; the output data is the context judgment value MW(q).
4. The method for identifying and removing cigarette debris based on images collected by an area array camera according to claim 1, characterized in that: Methods for identifying debris based on vein images include: Identify the vein defect area according to the vein image, extract the regional defect features corresponding to the vein defect area, determine whether it is debris according to the regional defect features, and obtain the debris identification result corresponding to the leaf vein area; Based on the leaf veins in the vein image, a simulation evaluation is performed on the non-vein area to determine whether there are debris and obtain the debris recognition result corresponding to the non-vein area.
Citation Information
Patent Citations
Forestry disease and pest intelligent identification method and system based on image understanding
CN115376032A
Tobacco leaf impurity identification method
CN116434011A
Baked tobacco leaf type detection method and system based on machine vision
CN116524224A
Intelligent photoelectric impurity removal method based on broken tobacco treatment
CN116965581A
Tobacco leaf impurity identification method and device, electronic equipment and storage medium
CN117197535A
