CN117853817B - Intelligent community garbage classification alarm management method based on image recognition - Google Patents
Intelligent community garbage classification alarm management method based on image recognition Download PDFInfo
- Publication number
- CN117853817B CN117853817B CN202410099046.7A CN202410099046A CN117853817B CN 117853817 B CN117853817 B CN 117853817B CN 202410099046 A CN202410099046 A CN 202410099046A CN 117853817 B CN117853817 B CN 117853817B
- Authority
- CN
- China
- Prior art keywords
- image
- garbage classification
- gradient
- information
- convolution
- 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
- 239000010813 municipal solid waste Substances 0.000 title claims abstract description 128
- 238000007726 management method Methods 0.000 title claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 51
- 238000010586 diagram Methods 0.000 claims abstract description 5
- 230000000694 effects Effects 0.000 claims description 39
- 238000000034 method Methods 0.000 claims description 37
- 230000006870 function Effects 0.000 claims description 32
- 230000008569 process Effects 0.000 claims description 25
- 238000006243 chemical reaction Methods 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 21
- 230000008859 change Effects 0.000 claims description 19
- 238000013136 deep learning model Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 11
- 230000005856 abnormality Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 8
- 238000010977 unit operation Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000001960 triggered effect Effects 0.000 claims 1
- 238000011156 evaluation Methods 0.000 abstract description 13
- 230000008447 perception Effects 0.000 abstract description 5
- 238000004458 analytical method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000012805 post-processing Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 230000003068 static effect Effects 0.000 description 4
- 239000002699 waste material Substances 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 230000006641 stabilisation Effects 0.000 description 3
- 238000011105 stabilization Methods 0.000 description 3
- 239000003086 colorant Substances 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 241000283973 Oryctolagus cuniculus Species 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000029087 digestion Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02W—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO WASTEWATER TREATMENT OR WASTE MANAGEMENT
- Y02W30/00—Technologies for solid waste management
- Y02W30/10—Waste collection, transportation, transfer or storage, e.g. segregated refuse collecting, electric or hybrid propulsion
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an intelligent community garbage classification alarm management method based on image recognition, and particularly relates to the field of garbage classification. The information programming coefficient is used as a comprehensive evaluation index, and the overall consistency and the local gradient stability are considered, so that the pertinence and the accuracy of basic image processing are improved. The evaluation result can accurately judge the availability of the basic data and prevent garbage classification errors. And a qualified characteristic diagram is adopted, a deformable convolution kernel is introduced to carry out convolution operation, and the adaptability to the shape and size changes of different garbage containers is improved. The deformable convolution network flexibly adjusts the convolution kernel, realizes accurate identification of the garbage container, adapts to various shapes, enhances the perception of a local structure, and improves the adaptability of the garbage classification system.
Description
Technical Field
The invention relates to the field of garbage classification, in particular to an intelligent community garbage classification alarm management method based on image recognition.
Background
Along with the continuous development of technology, intelligent garbage classification systems are deployed in communities for more effectively performing garbage classification digestion treatment, garbage images are analyzed and identified, garbage articles are classified into different categories, and intelligent garbage management is realized. However, the current garbage classification method based on image recognition has some problems in intelligent community management. Firstly, the technology may be affected by image quality, system hiding factors and aging problems, so that acquired basic image data and the acquired basic image data are reduced in the processing process, and subsequent recognition faults may be caused, so that garbage classification is interrupted. Secondly, for multi-class garbage classification in complex scenes, there is a limitation in adopting a static convolution network. Static convolution networks may perform poorly when dealing with garbage images of different sizes, shapes, and orientations, and are difficult to accommodate for the diverse image features. The poor adaptability of the dynamic changing garbage layout and environmental conditions in the static convolution check scene can lead to the performance reduction of the intelligent community garbage classification task with high requirements on real-time performance and complexity. In addition, the garbage classification method is sensitive to problems such as shielding and complex background, and the garbage classification accuracy in an actual scene is affected. The factors restrict the applicability of the static convolution network in dynamic and complex environments, challenge the intelligent community garbage classification management, and finally influence the timeliness of garbage classification alarm. In order to solve the problems, a new technical scheme is provided.
In order to solve the above problems, a technical solution is now provided.
Disclosure of Invention
In order to overcome the defects in the prior art, the embodiment of the invention provides an intelligent community garbage classification alarm management method based on image recognition, which is used for acquiring a community garbage classification scene image, extracting overall effect consistency and local gradient stability information by utilizing color space conversion information and image quality information, and obtaining an information programming coefficient by weighting the effect consistency index and the gradient stability index. The coefficient comprehensively and carefully evaluates the image processing quality and improves the sensitivity to the quality of the basic data. The information programming coefficient is used as a comprehensive evaluation index, and the overall consistency and the local gradient stability are considered, so that the pertinence and the accuracy of basic image processing are improved. The evaluation result can accurately judge the availability of the basic data and prevent garbage classification errors. And a qualified characteristic diagram is adopted, a deformable convolution kernel is introduced to carry out convolution operation, and the adaptability to the shape and size changes of different garbage containers is improved. The deformable convolution network flexibly adjusts the convolution kernel, realizes accurate identification of the garbage container, adapts to various shapes, enhances the perception of local structures, and improves the adaptability of the garbage classification system so as to solve the problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions: step S100, feature extraction information is carried out on the image processing process of garbage, and the preliminary processing reliability of the shot image data is judged;
Step S200, designing a deformable convolution network for carrying out convolution operation on the image aiming at the image which is judged to be qualified in advance, and introducing a deformable convolution kernel to adapt to the shape and size changes of different garbage containers;
Step S300, inputting the image extracted by the variability convolution feature into a deep learning model, constructing a convolution neural network, and training the deep learning model;
And step S400, outputting probability distribution through a model, judging abnormality according to a set threshold value, triggering early warning, and realizing garbage classification abnormality treatment.
In a preferred embodiment, step S100 specifically includes the following:
the method comprises the steps of obtaining real-time images of community garbage classification scenes by using a camera, extracting characteristic information according to the photographed images, wherein the characteristic information comprises color space conversion information and image quality information, the color space conversion information comprises an effect consistency index, and the image quality information comprises a gradient stability index.
In a preferred embodiment, the process of obtaining the effect consistency index is:
Step one, respectively calculating color histograms of photographed images before and after conversion, respectively converting the images before and after conversion into HSV color spaces, and respectively obtaining the color histograms of the photographed images before and after conversion;
Step two, calculating standard deviation for each channel of the images before and after conversion;
Step three, calculating information gain for each channel, and using differentiated color histogram and standard deviation change;
summing the information gains of all channels to obtain an overall information gain;
And fifthly, obtaining overall information gain for multiple times according to fixed acquisition interval time in unit operation time, and obtaining the effect consistency index through a standard deviation calculation mode.
In a preferred embodiment, the gradient stability index is obtained by:
Step one, converting a color image into a gray image;
step two, a gradient operator is used for convolving the gray level image, and gradients in the horizontal direction and the vertical direction are calculated;
Step three, calculating the amplitude value and the gradient direction of each pixel point and the gradient;
Step four, calculating the energy of the gradient amplitude and taking a square value;
Calculating the average value of the gradient energy to obtain the average gradient energy of the image;
and step six, obtaining average gradient energy for a plurality of times according to fixed acquisition interval time in unit operation time, and obtaining a gradient stability index through a standard deviation calculation mode.
In a preferred embodiment, the effect consistency index and the gradient stability index are weighted to obtain information programming coefficients;
After the information shielding coefficient is obtained, comparing the information shielding coefficient with a threshold value, and generating a qualified signal if the information shielding coefficient is larger than or equal to the threshold value; if the information programming coefficient is smaller than the threshold value, generating a hidden danger signal.
In a preferred embodiment, step S200 specifically includes the following:
step one, designing a dynamic convolution layer, comprising variability convolution kernels, and introducing a parameter theta= (theta x,θy) for each variability convolution kernel, wherein theta x and theta y respectively represent position offsets in x and y directions;
Step two, for each position (x, y) of the input feature map, calculating a corresponding position offset: Δx=Θ x*x;Δy=Θy x y;
Wherein, represents convolution operation;
step three, utilizing the position offset to generate sampling grid points (x+deltax, y+deltay) of the variability convolution kernel on the input feature map;
Step four, bilinear interpolation is carried out on sampling grid points, interpolation weights alpha and beta are obtained, and the interpolation weights alpha and beta are used for interpolating an input feature image:
Step five, weighting and superposing the input feature map by utilizing interpolation weight to obtain the output of the variability convolution kernel: y (x, y) = Σa, bαaβb·x (x+a, y+b);
wherein a and b are coordinates of the interpolation region;
in the training process, learning the position offset parameters of the variability convolution kernel through a back propagation algorithm;
The variability convolution operation is embedded into the whole deep learning model and combined with other layers to form a complete DCN network.
In a preferred embodiment, step S300 specifically includes the following:
Taking the image extracted by the variability convolution feature extracted in the step S200 as the input of a deep learning model;
Constructing a convolutional neural network, extracting high-level abstract feature representation by using a convolutional layer, and introducing nonlinearity through an activation function;
sampling on the input characteristic diagram by sliding a pooling core and fixing a stride;
applying a softmax activation function to obtain probability distribution of each category;
Model training is carried out by using marked garbage classification data, and the difference between model output and real labels is measured by adopting a cross entropy loss function;
updating model parameters by an optimizer by using a back propagation algorithm to minimize a loss function;
And performing multiple rounds of training until the value of the loss function is smaller than the corresponding threshold value, and further continuously improving the classification performance of the model.
In a preferred embodiment, step S400 specifically includes the following:
Acquiring probability distribution of the deep learning model after garbage classification of the image, namely the classification probability of each category;
And comparing each class probability output by the model with a set threshold value for each image, and if the probability of the class with the highest probability is smaller than the set threshold value or the difference between the probabilities of the two classes with the largest difference is larger than the set threshold value, judging the image as the abnormal condition of garbage classification and sending out an early warning signal.
The intelligent community garbage classification alarm management method based on image recognition has the technical effects and advantages that:
1. According to the invention, the community garbage classification scene image is obtained in real time through the camera, and the color space conversion information and the image quality information are extracted, so that the consistency evaluation of the overall effect of the image is obtained, and the local quality of the image is evaluated through the gradient stability index. And the availability of the basic data is facilitated to be comprehensively taken care of. And weighting the effect consistency index and the gradient stability index to obtain information programming coefficients, so that more comprehensive and fine evaluation indexes are provided for image processing, and the sensitivity to the quality of basic data is further improved. The information programming coefficient is used as a comprehensive evaluation index, so that the overall consistency of the image is considered, the stability of the local gradient is also considered, and the basic image processing is more targeted and accurate. The usability of the basic data can be judged more accurately through the evaluation of the information programming coefficient, so that errors caused by the quality problem of the data are found and avoided in advance in the subsequent processing process of garbage classification, and high-quality data support is provided for accurately classifying the garbage.
2. The invention can more flexibly adapt to the shape and size changes of different garbage containers by introducing the deformable convolution to carry out convolution operation on the image after the pre-judgment. The deformable convolution network realizes dynamic adjustment of the convolution kernel by learning the position deviation and the shape change of the target in the image, thereby improving the accurate identification of the garbage container. This flexibility is beneficial to cope with the diverse shapes of the trash receptacle in real scenes, making the system more adaptable and robust. Meanwhile, the introduction of the deformable convolution network can also effectively enhance the perception of local structures in the images, and further improve the overall performance of the garbage classification system, so that the garbage classification system is better suitable for garbage classification tasks in different community environments.
Drawings
FIG. 1 is a schematic diagram of a method for managing intelligent community garbage classification and alarm based on image recognition;
FIG. 2 is a flowchart for obtaining an effect consistency index of an intelligent community garbage classification alarm management method based on image recognition;
FIG. 3 is a flowchart of the acquisition of the gradient stability index of the intelligent community garbage classification alarm management method based on image recognition.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
FIG. 1 shows an intelligent community garbage classification alarm management method based on image recognition, which is characterized in that:
Step S100, feature extraction information is carried out on the image processing process of garbage, and the preliminary processing reliability of the shot image data is judged;
Step S200, designing a deformable convolution network for carrying out convolution operation on the image aiming at the image which is judged to be qualified in advance, and introducing a deformable convolution kernel to adapt to the shape and size changes of different garbage containers;
Step S300, inputting the image extracted by the variability convolution feature into a deep learning model, constructing a convolution neural network, and training the deep learning model;
And step S400, outputting probability distribution through a model, judging abnormality according to a set threshold value, triggering early warning, and realizing garbage classification abnormality treatment.
The step S100 specifically includes the following:
the method comprises the steps of obtaining real-time images of community garbage classification scenes by using a camera, extracting characteristic information according to the photographed images, wherein the characteristic information comprises color space conversion information and image quality information, the color space conversion information comprises an effect consistency index, and the image quality information comprises a gradient stability index.
As shown in fig. 2, the obtaining process of the effect consistency index is as follows:
The meaning of analyzing and calculating the effect consistency index in the aspect of earlier shot images is that the method provides an objective and comprehensive method for evaluating consistency and stability of shot images. In the garbage classification and identification system, the quality of the earlier shot image directly influences the later processing and identification effect. By measuring the consistency of the characteristics of gradients, colors and the like, the image obtained by the shooting system under different scenes and conditions can be ensured to have similar quality and characteristics. This is important to ensure consistency of the garbage classification system under different circumstances.
The meaning of the consistent image data for the later garbage classification and identification is mainly reflected in the aspect of improving the performance and stability of the model. In the machine learning model training process, consistent image data helps to provide more reliable training samples, enhancing the adaptability of the model to various scenes. If the consistency of the captured images is insufficient, the model may be affected by noise and variability, resulting in reduced performance. Therefore, by ensuring consistency of the captured images, we can increase the robustness of the garbage classification system, making it better suited for various capturing conditions, including different illumination, background and object sizes.
Analysis of the effect consistency index also helps to reduce post-processing uncertainty and adjustment complexity. If consistency of the photographed images is guaranteed, algorithms and parameters can be designed more robustly when image preprocessing, feature extraction and model training are performed. This helps reduce post-processing variables and interference factors, making the overall garbage classification system easier to maintain and upgrade. Furthermore, consistent image data helps to improve the repeatability of the system, making the image acquisition process more controllable at different points in time and places.
In general, the analysis and calculation effect consistency index has important significance for the garbage classification system. By ensuring the consistency of the earlier shot images, the model performance and the system stability can be improved, the uncertainty of later processing is reduced, and a foundation is laid for the reliable application of the garbage classification technology. The method not only helps to improve the automation level of the system, but also provides sustainability and maintainability support for long-term application of the system in a practical environment.
Step one, respectively calculating color histograms of photographed images before and after conversion, respectively converting the images I before and I after into HSV color spaces, and respectively obtaining color histograms H before and H after of the images;
step two, calculating standard deviation of each channel of the images before and after conversion, and setting the standard deviation of the o-th channel as And
Step three, calculating information gain for each channel o, and using the differentiated color histogram and standard deviation change, wherein the calculation formula of the information gain is as follows:
where ε is a smoothing factor and the denominator is avoided being zero.
Step four, summing the information gains of all channels to obtain the whole information gain, namely
The calculation formula is as follows:
Where N is the number of channels.
And fifthly, obtaining overall information gain for multiple times according to fixed acquisition interval time in unit operation time, and obtaining the effect consistency index through a standard deviation calculation mode.
The effect coincidence index is used to represent the degree of coincidence of images before and after color space conversion. When the effect consistency index is larger, the difference of the converted image in the color space is larger, namely the color information is obviously changed, the fluctuation of the information difference between the original image obtained through shooting and the processed image is larger, the fluctuation of the effect difference before and after the image processing is larger, and the fluctuation of hidden danger factors which are unstable in the aspect of basic data processing is shown. Conversely, when the effect coincidence index is smaller, it means that the converted image is more coincident in color space with the image before conversion, and the change in color information is relatively small. This means that the fluctuation of the information difference between the photographed original image and the processed image is small, and the difference in effect before and after the image processing is relatively stable. In the application of the garbage classification system, the stability shows that the color space conversion does not introduce excessive changes, the system has relatively stable performance in the aspect of basic data processing, random fluctuation caused by image processing is reduced, and the robustness and reliability of the system are improved. Therefore, the magnitude of the effect consistency index reflects the variation stability of the effects before and after image processing, and provides an important reference for performance evaluation of the system.
As shown in fig. 3, the gradient stability index is obtained by:
The significance of analyzing and calculating the gradient stability index in the aspect of earlier shooting images is mainly that a powerful tool is provided for evaluating the stability and consistency of a shooting system on the gradient characteristics of the images. In the garbage classification and identification system, the quality and stability of the earlier shot images directly influence the accuracy of later processing and identification. By calculating the gradient stability index, the gradient change degree of the photographed image can be quantitatively measured, so that images obtained under different scenes and conditions are ensured to have similar gradient characteristics, and the consistency of image data is ensured.
The importance of the gradient stabilization index in the analysis of pre-captured images is first manifested in its objective assessment of image quality. Gradient features are important representations of details such as edges and textures in images, which are critical to garbage classification recognition. By analyzing the gradient stability index, the degree of the image gradient retention of the shooting system under different scenes can be known, and further the stability of the image quality can be judged. The consistent gradient characteristics are helpful for ensuring the definition and the information richness of the images and improving the preliminary understanding of the garbage classification system on the images.
The significance of the post garbage classification and identification is mainly reflected in the aspect of improving the performance and stability of the model. The analysis of the gradient stability index may provide more robust training data for the machine learning model. The consistent gradient features help the model to better learn and understand the similarity between different images, enhancing the adaptability of the model to various scenes. This is critical to the popularization and practical application of the garbage classification system, because the shooting conditions in reality may be various, including illumination changes, background interference, etc. By ensuring the stability of the image gradient, the robustness of the model can be improved, so that the model can better cope with changeable shooting environments.
Furthermore, analysis of the gradient stabilization index helps to reduce post-processing uncertainty and complexity of adjustment. If the gradient change of the photographed image is unstable, the later image processing and feature extraction may be disturbed by noise, resulting in degradation of recognition performance. By ensuring gradient stability, the image processing algorithm and model parameters can be designed more robustly, and variables and interference factors of post-processing are reduced, so that the whole garbage classification system is easier to maintain and upgrade.
Overall, analysis of the gradient stability index is of great significance to the garbage classification system. By ensuring the gradient stability of the earlier shot images, the model performance and the system stability can be improved, the uncertainty of later treatment is reduced, and a foundation is laid for the reliable application of the garbage classification technology. The method not only helps to improve the automation level of the system, but also provides sustainability and maintainability support for long-term application of the system in a practical environment.
Step one, converting a color image into a gray image so as to simplify the processing process;
the calculation formula is as follows:
Igray(x,y)=0.299·Irgb(x,y)R+0.587·Irgb(x,y)G+0.114·Irgb(x,y)B
The formula is a conversion formula from a color image to a gray image, and a weighted average value is used to reflect the perception weights of human eyes to different colors. The meaning of the processing is that the image processing flow is simplified, the color information is converted into brightness information, and the subsequent processing is more convenient.
Step two, a gradient operator is used for convolving the gray level image, and gradients in the horizontal direction and the vertical direction are calculated;
The formula:
S x and S y are convolution kernels in the horizontal and vertical directions of the Sobel operator, and gradient of each pixel point in the image is calculated through convolution operation, and the gradient represents changing intensity in the horizontal and vertical directions respectively, so that contours and edges in the image can be captured;
Step three, calculating the amplitude value and the gradient direction of each pixel point and the gradient;
The formula:
GradientDirection(x,y)=arctan2(Gy(x,y),Gx(x,y));
Respectively calculating the gradient amplitude and the gradient direction of each pixel point, wherein the gradient amplitude represents the change intensity of the pixel, and the gradient direction represents the change direction, which are very important for measuring the details and the structural information in the image;
Step four, calculating the energy of the gradient amplitude and taking a square value;
The formula:
Energy(x,y)=GradientMagnitude(x,y)2;
Calculating the energy of the gradient amplitude, selecting the square, and helping to highlight the change intensity of the gradient amplitude, wherein the larger the value is, the more severe the change is, and more edges and textures exist in the image, so that important details in the image are easier to capture;
And fifthly, calculating the average value of the gradient energy to obtain the average gradient energy of the image.
And step six, obtaining average gradient energy for a plurality of times according to fixed acquisition interval time in unit operation time, and obtaining a gradient stability index through a standard deviation calculation mode.
The gradient stability index is used for reflecting the stability and consistency of gradient energy in the image. Specifically, the gradient stability index reflects the fluctuation of the gradient characteristics of the image within a certain time range by calculating the average energy value of the gradient and the standard deviation thereof. The stability of the gradient energy is critical for image processing and model training.
When the gradient stabilization index is larger, the fluctuation of gradient energy in the image is larger, and the gradient characteristics are relatively unstable. This may indicate that the image quality is not consistent enough, the gradient information changes greatly at different sampling time points, which may lead to an increase in uncertainty of subsequent processing and model training, which indicates that the difference between effects before and after image processing fluctuates greatly, and indicates that the system has unstable hidden trouble factor fluctuation in the aspect of basic data processing. Conversely, a smaller gradient stability index indicates that the gradient energy is relatively stable and that the gradient characteristics of the image at different points in time are relatively consistent. This helps to improve consistency of image processing, enhances adaptation of the model to various scenarios, and reduces post-processing uncertainty. Thus, a smaller gradient stability index is generally more beneficial to the performance and stability of the system.
The information programming coefficient is obtained by weighting the effect consistency index and the gradient stability index, for example, the information programming coefficient can be obtained by the following calculation formula: c=ln [ w1·gt+w2·et+1];
In the formula, C represents an information programming coefficient, GT and ET respectively represent an effect consistency index and a gradient stability index, and w1 and w2 respectively represent preset proportionality coefficients of the effect consistency index and the gradient stability index, and are both larger than zero.
The information programming coefficient represents the overall quality and stability of the data processing. The larger information programming coefficient indicates that the basic image data processing has higher consistency and stability, namely, the basic image data is relatively consistent under different situations, and the gradient characteristics are relatively stable. This is critical to the performance enhancement of subsequent image processing and garbage classification models. Conversely, smaller information-encoding coefficients mean poorer consistency and stability of data processing, and the base image is prone to introduce problems of uncertainty, which can present challenges for the accuracy and robustness of the later model. Therefore, the size of the information programming coefficient directly reflects the consistency and stability level in the processing process of the basic image data, and is one of key marks for evaluating the processing effect of the whole garbage classification and identification system.
After the information programming coefficient is obtained, the information programming coefficient is compared with a threshold value, and if the information programming coefficient is larger than or equal to the threshold value, the consistency and the stability of the basic picture data processing reach the expected level. Specifically, the effect consistency and gradient stability in the system are high enough to meet the set threshold requirements, so that powerful support is provided for classifying garbage in the later period, subsequent tasks such as image processing and garbage classification are facilitated, and qualified signals are generated.
If the information programming factor is less than the threshold, it indicates that the consistency and stability of the data processing has not reached the desired level. The effect consistency and gradient stability of processing the pictures photographed by the garbage are relatively poor, the condition of fluctuation or inconsistency of processing quality exists, the situation of low recognition accuracy and the like is easy to cause for the garbage classification in the later period, the hidden danger signal is generated, the follow-up processing flow is not participated, and meanwhile, the early warning prompt is sent out to update, reform and maintain.
According to the invention, the community garbage classification scene image is obtained in real time through the camera, and the color space conversion information and the image quality information are extracted, so that the consistency evaluation of the overall effect of the image is obtained, and the local quality of the image is evaluated through the gradient stability index. And the availability of the basic data is facilitated to be comprehensively taken care of. And weighting the effect consistency index and the gradient stability index to obtain information programming coefficients, so that more comprehensive and fine evaluation indexes are provided for image processing, and the sensitivity to the quality of basic data is further improved. The information programming coefficient is used as a comprehensive evaluation index, so that the overall consistency of the image is considered, the stability of the local gradient is also considered, and the basic image processing is more targeted and accurate. The usability of the basic data can be judged more accurately through the evaluation of the information programming coefficient, so that errors caused by the quality problem of the data are found and avoided in advance in the subsequent processing process of garbage classification, and high-quality data support is provided for accurately classifying the garbage.
The step S200 specifically includes the following:
The design of the variable convolution layer has the significance of improving the adaptability of the Convolution Neural Network (CNN) to the shape and size changes of different garbage containers, so that the feature extraction capability of the garbage image is enhanced. Conventional convolution layers use a fixed convolution kernel, which may perform poorly when dealing with different shapes and sizes of garbage containers. The variability convolution layer introduces the concept of a deformable convolution kernel, allowing the convolution kernel to dynamically adjust shape according to local features of the input image, adapting to variability of the garbage container. This dynamics enables the network to better capture detailed information in the spam image, especially in complex scenarios, such as where different shapes and sizes of spam containers are mixed. Through designing the variability convolution layer, the network can adapt to the diversity of rubbish image in the actual environment more flexibly, improves the recognition accuracy to the rubbish container of different shapes and sizes, provides stronger feature extraction ability for the deep learning model, and then promotes whole intelligent community rubbish classification system's performance and stability. The innovative design solves the limitation of the traditional convolution layer in coping with the diversity of the garbage images, and brings new breakthrough to the development of garbage classification technology.
Step one, designing a dynamic convolution layer, comprising variability convolution kernels, and introducing a parameter theta= (theta x,θy) for each variability convolution kernel, wherein theta x and theta y respectively represent position offsets in x and y directions;
Step two, for each position (x, y) of the input feature map, calculating a corresponding position offset: Δx=Θ x*x;Δy=Θy x y;
where x represents the convolution operation.
The input feature map here refers to an image for which it is judged that the pass signal is obtained through step S100.
Step three, utilizing the position offset to generate sampling grid points (x+deltax, y+deltay) of the variability convolution kernel on the input feature map;
The sampling position of the convolution kernel is dynamically determined on the input feature map, so that the convolution kernel can sample in a region of interest, the convolution kernel can be more flexible and is not limited by a fixed receptive field, the perceptibility of the model to the change of the target position and the shape is improved, and the representation capability of the image features is enhanced. In a waste sorting system, this helps to better identify waste containers of different shapes and locations.
Step four, bilinear interpolation is carried out on sampling grid points, interpolation weights alpha and beta are obtained, and the interpolation weights alpha and beta are used for interpolating an input feature image:
Interpolation weights α and β are used to weight overlap between four adjacent pixels of the interpolation region to calculate the output of the variability convolution kernel at the sampling point. Through the interpolation operation, the pixel values at different positions can be fused more continuously, and the precision of the variability convolution operation is improved.
Step five, weighting and superposing the input feature map by utilizing interpolation weight to obtain the output of the variability convolution kernel: y (x, y) = Σ a,bαaβb ·x (x+a, y+b);
where a and b are the coordinates of the interpolation region.
The input feature images are weighted and overlapped through interpolation weights, and output of a variability convolution kernel is generated, so that the method is an operation for integrating feature information of different positions. The interpolation weights control the degree of contribution to neighboring pixel features, thereby more flexibly capturing the position and shape changes of the object in the output of the variability convolution kernel. The effect of this operation is to enhance the local perceptibility of the input image by the network, enabling the variability convolution kernel to adapt more accurately to the characteristics of the target at different locations. Therefore, the output of the variability convolution kernel is more adaptive through the weighted superposition of interpolation weights, and the recognition accuracy of the model to different target shapes and positions is improved.
In the training process, the position offset parameters of the variability convolution kernel are learned through a back propagation algorithm, so that the network can automatically adapt to the change of the target shape. The network can learn the change rule of the target shape from the data, the generalization capability of the model is improved, and the adaptability to different shapes of the garbage container is enhanced.
The specific treatment process is as follows:
1. A penalty function is defined, including classification penalty and regularization term. Let L cls be the classification loss and L reg be the regularization term, the total loss L can be expressed as: l=l cls+λ·Lreg;
where λ is the weight of the regularization term.
2. The difference between the model output and the real label is calculated using a classification loss function, such as cross entropy loss: l cls=-∑cyclog(pc);
Where y c is the probability distribution of the real labels and p c is the probability distribution of the model predictions.
3. Using a back propagation algorithm, the gradient of the loss function over the position offset parameter Θ= (θ x,θy) is calculated:
Wherein, Representing the gradient of the classification loss to the output y (p), whileRepresenting the gradient of the output y (p) to the positional offset parameter θ x. The gradient of the classification loss is propagated to the positional shift parameter by the chain law.
Also, the process of the present invention is,Representing the gradient of the output y (p) to the position offset parameter θ y, z and s represent the coordinates of the convolution kernel.
These gradients can be automatically calculated by a back propagation algorithm. By adjusting the position deviation parameters, the model can learn to adapt to the change of the shape of the target, so that the adaptability to targets with different shapes is improved.
4. Updating the position offset parameter by a gradient descent rule:
where k is the learning rate and where k is the learning rate,
5. The regularization term is used to control the magnitude of the positional offset parameter to prevent overfitting. A common regularization term may be an L1 or L2 norm:
6. Combining the classification loss and the gradient of the regularization term, and updating parameters of the model:
7. updating the position offset parameters of the variability convolution kernel according to the overall gradient:
8. Repeating the steps, and gradually learning the variable convolution kernel parameters adapting to the shape change of the target by the network through repeated iterative training.
Through the 1-8 process, the position deviation parameter of the variability convolution kernel can be automatically adjusted according to the change of the target shape in the training data, and the adaptability of the network to targets with different shapes is improved.
Appropriate initial values for the position offset and shape change parameters are set to encourage the model to learn the valid parameters faster.
The number of variability convolution kernels is determined based on the complexity of the task and the availability of computing resources. Multiple variability convolution kernels may be used in the network to increase flexibility.
The variability convolution operation is embedded into the whole deep learning model and combined with other layers to form a complete DCN network, and the integration of the DCN network enables the whole model to have stronger feature extraction capability, so that garbage containers with different shapes and sizes can be better captured and identified in garbage classification scenes.
The invention can more flexibly adapt to the shape and size changes of different garbage containers by introducing the deformable convolution to carry out convolution operation on the image after the pre-judgment. The deformable convolution network realizes dynamic adjustment of the convolution kernel by learning the position deviation and the shape change of the target in the image, thereby improving the accurate identification of the garbage container. This flexibility is beneficial to cope with the diverse shapes of the trash receptacle in real scenes, making the system more adaptable and robust. Meanwhile, the introduction of the deformable convolution network can also effectively enhance the perception of local structures in the images, and further improve the overall performance of the garbage classification system, so that the garbage classification system is better suitable for garbage classification tasks in different community environments.
The step S300 specifically includes the following:
Taking the image extracted by the variability convolution feature extracted in the step S200 as the input of a deep learning model;
constructing a convolutional neural network, wherein the convolutional neural network comprises a convolutional layer, a pooling layer, a full-connection layer and the like;
Extracting high-level abstract feature representation by using a convolution layer, and introducing nonlinearity through an activation function;
assuming a convolution kernel and an input feature map, the calculation process of the convolution operation is as follows:
and sliding the convolution kernel on the belonging rabbit feature map, multiplying the elements at the corresponding positions one by one, and adding all the product results to obtain the output of the convolution operation.
The calculation formula is as follows:
Wherein W represents a convolution kernel, X represents a feature map, Z represents the output of the convolution operation, M and N are the height and width of the convolution kernel, respectively, and X (r+m, t+n). W (M, N) is the pixel value at the corresponding position on the input feature.
The output of the convolution operation is then added to the paranoid term.
The calculation formula is as follows: z (r, t) =z (r, t) +h;
where h represents a paranoid.
And inputting the result obtained by the convolution operation and the addition of the offset term into an activation function, and introducing nonlinearity.
The calculation formula is as follows: a (r, t) =σ (Z (r, t));
Where a (r, t) is the activation value of the corresponding position on the feature map output by the convolutional layer, σ represents the activation function, r, t represent the spatial position coordinates on the feature map, specifically r represents the row coordinates, and t represents the column coordinates, respectively.
Common activation functions include ReLU (RectifiedLinearUnit), sigmoid, tanh, etc., which can introduce nonlinearities in different situations so that the network can learn more complex feature representations. For example, the ReLU activation function is calculated by ReLU (x) =max (0, x), with negative values set to zero, and positive values retained. Thus, through the processing of the convolution layer, the network can learn and extract abstract features in the input image, and a more meaningful representation is provided for the subsequent garbage classification task.
Downsampling is carried out by using a pooling layer, the dimension of the feature map is reduced, and important information is reserved;
assuming that a pooling core (poolingkernel) has a size of f×f, the calculation process of the pooling layer is as follows: inputting the pooled features into a full-connection layer, and mapping the image features to garbage classification categories;
And sliding the pooling core on the input feature map, and taking the maximum value in the element selection window at the corresponding position as the output of the pooling operation.
The calculation formula is as follows:
Where Y (r, t) is the output of the pooling operation and S is the stride of the pooling operation.
And sliding the pooling core on the input feature map, and taking an average value in a window as the output of pooling operation by the element at the corresponding position.
The calculation formula is as follows:
Wherein Y (r, t) is the value of the corresponding position on the feature map of the chemical layer output.
The pooling operation samples the input feature map with a fixed stride through sliding the pooling core, reduces the size of the feature map, and selects the maximum value or average value in the window at the same time, thereby realizing the simplification of the image information. This helps to reduce computational complexity, reduce the number of parameters of the model, while preserving the main features in the image, providing more focused information for subsequent fully connected layers.
Applying a softmax activation function to obtain probability distribution of each category;
Let the output of the neural network be a vector z= [ z 1,z2,...,zC ] containing C elements, where C is the number of categories.
The Softmax activation function is calculated as follows:
calculating an unnormalized index value (exponentials):
Calculating the sum of the index values:
calculating probability distribution of each category:
where P (y u = 1|z) represents the probability that the sample belongs to the u-th class given the input z.
In this way, the Softmax activation function ensures that the probability distribution of the output satisfies the nature of the probability, i.e. the probability sum is 1. In the garbage classification problem, after the output of the model goes through the Softmax activation function, each element represents the probability that the image belongs to the corresponding class. The model will eventually predict as the category with the highest probability.
Model training is carried out by using marked garbage classification data, and the difference between model output and real labels is measured by adopting a cross entropy loss function;
Assuming that there are C classes, the output of the model is a probability distribution p= [ P 1,P2,...,PC ], and the true label is in the form y= [ Y 1,Y2,...,YC ] of one-hot encoding, where Y f represents the true label that the sample belongs to the f-th class.
The cross entropy loss function is calculated as follows:
The specific calculation steps are as follows:
1. for each class f, the cross entropy loss for that class is calculated:
lossf=-Yf·log(Pf);
2. adding the losses of each category to obtain the overall cross entropy loss:
Through the loss function, the object of the model in the training process is to reduce cross entropy loss as much as possible, so that the accuracy of the model on garbage classification tasks is improved. The optimizer (such as random gradient descent) updates the model parameters using a back propagation algorithm so that the predicted probability distribution of the model is closer to the real label, and finally a better classification effect is achieved.
Updating model parameters by an optimizer by using a back propagation algorithm to minimize a loss function;
For each model parameter θ, the partial derivative of the loss function with respect to that parameter, i.e., the gradient, is calculated. The gradient is propagated from the loss function to the model parameters using the chain law.
The calculation formula is as follows:
model parameters are updated with optimizers (e.g., random gradient descent) based on gradient information.
Updating the formula:
where η is the learning rate, the step size of each parameter update is controlled.
And performing multiple rounds of training until the value of the loss function is smaller than the corresponding threshold value, and further continuously improving the classification performance of the model.
Step S400 specifically includes the following:
Acquiring probability distribution of the deep learning model after garbage classification of the image, namely the classification probability of each category;
And comparing each class probability output by the model with a set threshold value for each image, judging the image as the abnormal condition of garbage classification if the probability of the class with the highest probability is smaller than the set threshold value or the difference between the probabilities of the two classes with the largest difference is larger than the set threshold value, sending out an early warning signal, and transferring the garbage with disputes to the manual sorting area.
First, obtaining classification probabilities for respective categories can provide detailed garbage classification information. This allows the system to determine not only the types of spam present in the image, but also to understand the confidence of the model in classifying different categories. Such probability distribution information is helpful for interpretation and visualization of garbage classification results, and transparency and understandability of the system are improved. Secondly, by setting a threshold value to perform abnormality detection, the system can automatically judge whether the garbage classification abnormality exists. When the highest probability class probability is lower than a set threshold value or the difference between the two maximum class probabilities is greater than the set threshold value, the system triggers an early warning signal. The method effectively reduces the false alarm rate, and gives an early warning only when the model is uncertain or ambiguous in garbage classification, thereby improving the accuracy and reliability of the system. Finally, once the garbage classification abnormality is found, the system can take timely measures, such as sending out early warning signals and transferring the garbage with the early warning signals to a manual sorting area. This helps to avoid misclassified waste entering the wrong process flow, ensuring the overall efficiency of the waste classification system. The system integrates the output of the deep learning model and the abnormality detection mechanism of threshold setting, and can be more intelligent and self-adaptive in the garbage classification process, so that the community management efficiency is improved.
The above formulas are all formulas with dimensionality removed and numerical calculation, the formulas are formulas with the latest real situation obtained by software simulation through collecting a large amount of data, and preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.
The above embodiments may be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented in software, the above-described embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with embodiments of the present application are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more sets of available media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium may be a solid state disk.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
In the several embodiments provided in the present application, it should be understood that the disclosed method and apparatus may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other forms.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, or the like) to perform all or part of the steps described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, or other various media capable of storing program codes.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Finally: the foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (4)
1. An intelligent community garbage classification alarm management method based on image recognition is characterized by comprising the following steps of:
Step S100, feature extraction information is carried out on the image processing process of garbage, and the preliminary processing reliability of the shot image data is judged;
Step S200, designing a deformable convolution network for carrying out convolution operation on the image aiming at the image which is judged to be qualified in advance, and introducing a deformable convolution kernel to adapt to the shape and size changes of different garbage containers;
Step S300, inputting the image extracted by the variability convolution feature into a deep learning model, constructing a convolution neural network, and training the deep learning model;
step S400, probability distribution is output through a model, abnormality is judged according to a set threshold value, early warning is triggered, and garbage classification abnormality processing is realized;
The step S100 specifically includes the following:
acquiring real-time images of community garbage classification scenes by using a camera, extracting characteristic information according to the shot images, wherein the characteristic information comprises color space conversion information and image quality information, the color space conversion information comprises an effect consistency index, and the image quality information comprises a gradient stability index;
the obtaining process of the effect consistency index comprises the following steps:
Step one, respectively calculating color histograms of photographed images before and after conversion, respectively converting the images before and after conversion into HSV color spaces, and respectively obtaining the color histograms of the photographed images before and after conversion;
Step two, calculating standard deviation for each channel of the images before and after conversion;
Step three, calculating information gain for each channel, and using differentiated color histogram and standard deviation change;
summing the information gains of all channels to obtain an overall information gain;
Step five, obtaining overall information gain for multiple times according to fixed acquisition interval time in unit operation time, and obtaining an effect consistency index through a standard deviation calculation mode;
the gradient stability index is obtained by the following steps:
Step one, converting a color image into a gray image;
step two, a gradient operator is used for convolving the gray level image, and gradients in the horizontal direction and the vertical direction are calculated;
Step three, calculating the amplitude value and the gradient direction of each pixel point and the gradient;
Step four, calculating the energy of the gradient amplitude and taking a square value;
Calculating the average value of the gradient energy to obtain the average gradient energy of the image;
Step six, obtaining average gradient energy for a plurality of times according to fixed acquisition interval time in unit operation time, and obtaining a gradient stability index through a standard deviation calculation mode;
The information programming coefficient is obtained after the effect consistency index and the gradient stability index are weighted;
After the information shielding coefficient is obtained, comparing the information shielding coefficient with a threshold value, and generating a qualified signal if the information shielding coefficient is larger than or equal to the threshold value; if the information programming coefficient is smaller than the threshold value, generating a hidden danger signal.
2. The intelligent community garbage classification alarm management method based on image recognition according to claim 1, wherein the intelligent community garbage classification alarm management method based on image recognition is characterized by comprising the following steps:
The step S200 specifically includes the following:
step one, designing a dynamic convolution layer, comprising variability convolution kernels, and introducing a parameter theta= (theta x,θy) for each variability convolution kernel, wherein theta x and theta y respectively represent position offsets in x and y directions;
Step two, for each position (x, y) of the input feature map, calculating a corresponding position offset: Δx=Θ x*x;Δy=Θy x y;
Wherein, represents convolution operation;
step three, utilizing the position offset to generate sampling grid points (x+deltax, y+deltay) of the variability convolution kernel on the input feature map;
Step four, bilinear interpolation is carried out on sampling grid points, interpolation weights alpha and beta are obtained, and the interpolation weights alpha and beta are used for interpolating an input feature image:
Step five, weighting and superposing the input feature map by utilizing interpolation weight to obtain the output of the variability convolution kernel: y (x, y) = Σ a,bαaβb ·x (x+a, y+b);
wherein a and b are coordinates of the interpolation region;
in the training process, learning the position offset parameters of the variability convolution kernel through a back propagation algorithm;
The variability convolution operation is embedded into the whole deep learning model and combined with other layers to form a complete DCN network.
3. The intelligent community garbage classification alarm management method based on image recognition according to claim 2, wherein the intelligent community garbage classification alarm management method based on image recognition is characterized in that:
the step S300 specifically includes the following:
Taking the image extracted by the variability convolution feature extracted in the step S200 as the input of a deep learning model;
Constructing a convolutional neural network, extracting high-level abstract feature representation by using a convolutional layer, and introducing nonlinearity through an activation function;
sampling on the input characteristic diagram by sliding a pooling core and fixing a stride;
applying a softmax activation function to obtain probability distribution of each category;
Model training is carried out by using marked garbage classification data, and the difference between model output and real labels is measured by adopting a cross entropy loss function;
updating model parameters by an optimizer by using a back propagation algorithm to minimize a loss function;
multiple rounds of training are performed until the value of the loss function is less than the corresponding threshold.
4. The intelligent community garbage classification alarm management method based on image recognition according to claim 3, wherein the intelligent community garbage classification alarm management method based on image recognition is characterized by comprising the following steps of:
Step S400 specifically includes the following:
Acquiring probability distribution of the deep learning model after garbage classification of the image, namely the classification probability of each category;
And comparing each class probability output by the model with a set threshold value for each image, and if the probability of the class with the highest probability is smaller than the set threshold value or the difference between the probabilities of the two classes with the largest difference is larger than the set threshold value, judging the image as the abnormal condition of garbage classification and sending out an early warning signal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410099046.7A CN117853817B (en) | 2024-01-24 | 2024-01-24 | Intelligent community garbage classification alarm management method based on image recognition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410099046.7A CN117853817B (en) | 2024-01-24 | 2024-01-24 | Intelligent community garbage classification alarm management method based on image recognition |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117853817A CN117853817A (en) | 2024-04-09 |
CN117853817B true CN117853817B (en) | 2024-06-04 |
Family
ID=90541769
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410099046.7A Active CN117853817B (en) | 2024-01-24 | 2024-01-24 | Intelligent community garbage classification alarm management method based on image recognition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117853817B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108734102A (en) * | 2018-04-18 | 2018-11-02 | 佛山市顺德区中山大学研究院 | A kind of right and left eyes recognizer based on deep learning |
CN109325417A (en) * | 2018-08-23 | 2019-02-12 | 东北大学 | A kind of industrial process fault condition diagnostic method based on deep neural network |
CN109345491A (en) * | 2018-09-26 | 2019-02-15 | 中国科学院西安光学精密机械研究所 | Remote sensing image enhancement method fusing gradient and gray scale information |
WO2019144581A1 (en) * | 2018-01-29 | 2019-08-01 | 江苏宇特光电科技股份有限公司 | Smart infrared image scene enhancement method |
WO2020037965A1 (en) * | 2018-08-21 | 2020-02-27 | 北京大学深圳研究生院 | Method for multi-motion flow deep convolutional network model for video prediction |
CN111079639A (en) * | 2019-12-13 | 2020-04-28 | 中国平安财产保险股份有限公司 | Method, device and equipment for constructing garbage image classification model and storage medium |
CN111767804A (en) * | 2020-06-08 | 2020-10-13 | 上海交通大学 | Recyclable garbage image classification method and system based on artificial intelligence |
JP6980958B1 (en) * | 2021-06-23 | 2021-12-15 | 中国科学院西北生態環境資源研究院 | Rural area classification garbage identification method based on deep learning |
CN116758357A (en) * | 2023-07-17 | 2023-09-15 | 青岛北海俊惠电子仪表有限公司 | Automatic identification method for pipe-rod coupling of oil field |
CN117409339A (en) * | 2023-10-13 | 2024-01-16 | 东南大学 | Unmanned aerial vehicle crop state visual identification method for air-ground coordination |
-
2024
- 2024-01-24 CN CN202410099046.7A patent/CN117853817B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019144581A1 (en) * | 2018-01-29 | 2019-08-01 | 江苏宇特光电科技股份有限公司 | Smart infrared image scene enhancement method |
CN108734102A (en) * | 2018-04-18 | 2018-11-02 | 佛山市顺德区中山大学研究院 | A kind of right and left eyes recognizer based on deep learning |
WO2020037965A1 (en) * | 2018-08-21 | 2020-02-27 | 北京大学深圳研究生院 | Method for multi-motion flow deep convolutional network model for video prediction |
CN109325417A (en) * | 2018-08-23 | 2019-02-12 | 东北大学 | A kind of industrial process fault condition diagnostic method based on deep neural network |
CN109345491A (en) * | 2018-09-26 | 2019-02-15 | 中国科学院西安光学精密机械研究所 | Remote sensing image enhancement method fusing gradient and gray scale information |
CN111079639A (en) * | 2019-12-13 | 2020-04-28 | 中国平安财产保险股份有限公司 | Method, device and equipment for constructing garbage image classification model and storage medium |
CN111767804A (en) * | 2020-06-08 | 2020-10-13 | 上海交通大学 | Recyclable garbage image classification method and system based on artificial intelligence |
JP6980958B1 (en) * | 2021-06-23 | 2021-12-15 | 中国科学院西北生態環境資源研究院 | Rural area classification garbage identification method based on deep learning |
CN116758357A (en) * | 2023-07-17 | 2023-09-15 | 青岛北海俊惠电子仪表有限公司 | Automatic identification method for pipe-rod coupling of oil field |
CN117409339A (en) * | 2023-10-13 | 2024-01-16 | 东南大学 | Unmanned aerial vehicle crop state visual identification method for air-ground coordination |
Non-Patent Citations (9)
Title |
---|
Grigoryan, Artyom M.Gradient Based Histogram Equalization in Grayscale Image Enhancement.《MOBILE MULTIMEDIA/IMAGE PROCESSING, SECURITY, AND APPLICATIONS》.2019,全文. * |
Iulia–Nela Anghelache Nastase等.Improved Classification of Benign and Malignant Breast Lesions Based on Contrast Manipulation and Gradient Map.《2022 E-Health and Bioengineering Conference (EHB)》.2022,全文. * |
Power Line Recognition and Foreign Objects Detection Based on Image Processing;Zhenlin Song等;《2021 33rd Chinese Control and Decision Conference》;20211130;全文 * |
升余弦变增益微视觉图像自适应增强与应用;张翔;张宪民;李海;李凯;;光学精密工程;20180515(第05期);全文 * |
张振宇 ; 朱培栋 ; 王可 ; 胡慧俐 ; .拓扑结构与节点属性综合分析的社区发现算法.计算机技术与发展.2017,(第04期),全文. * |
李梦天.基于深度学习的高效渲染技术研究.《硕士电子期刊出版信息》.2021,全文. * |
石祥滨 ; 阳柯 ; 张德园 ; .基于对称神经网络的图像分类.小型微型计算机系统.2017,(第03期),全文. * |
秦斌斌 ; 何级 ; .基于卷积神经网络的垃圾分类研究.无线通信技术.2019,(第03期),全文. * |
赵若晴 ; 王慧琴 ; 王可 ; 王展 ; 刘文腾 ; .基于方向梯度直方图和灰度共生矩阵混合特征的金文图像识别.激光与光电子学进展.2020,(第12期),全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN117853817A (en) | 2024-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
JP5546317B2 (en) | Visual inspection device, visual inspection discriminator generation device, visual inspection discriminator generation method, and visual inspection discriminator generation computer program | |
CN113298058A (en) | Water quality prediction inversion method and system based on hyperspectral image of unmanned aerial vehicle | |
CN113592060A (en) | Neural network optimization method and device | |
CN111832228A (en) | Vibration transmission system based on CNN-LSTM | |
CN113011398A (en) | Target change detection method and device for multi-temporal remote sensing image | |
CN115965862A (en) | SAR ship target detection method based on mask network fusion image characteristics | |
CN116883933A (en) | Security inspection contraband detection method based on multi-scale attention and data enhancement | |
CN111524113A (en) | Lifting chain abnormity identification method, system, equipment and medium | |
CN117576724A (en) | Unmanned plane bird detection method, system, equipment and medium | |
Wang et al. | Classification and extent determination of rock slope using deep learning | |
Yang et al. | Rapid image detection and recognition of rice false smut based on mobile smart devices with anti-light features from cloud database | |
CN117994240A (en) | Multi-scale two-level optical remote sensing image stripe noise intelligent detection method and device | |
CN113378642A (en) | Method for detecting illegal occupation buildings in rural areas | |
CN117853817B (en) | Intelligent community garbage classification alarm management method based on image recognition | |
CN115984712A (en) | Multi-scale feature-based remote sensing image small target detection method and system | |
CN112396648B (en) | Target identification method and system capable of positioning mass center of target object | |
CN112434710A (en) | Intelligent goods shelf goods identification method based on target detection technology | |
CN117689880B (en) | Method and system for target recognition in biomedical images based on machine learning | |
CN117809169B (en) | Small-sample underwater sonar image classification method and model building method thereof | |
Nirmaladevi et al. | Prediction of leaf disease and pest detection using deep learning | |
CN117953349B (en) | Method, device, equipment and storage medium for detecting plant diseases and insect pests of traditional Chinese medicinal materials | |
CN118570569B (en) | Mosquito identification method and system based on convolutional neural network | |
KR102418476B1 (en) | Method of operating parking control system that precisely performs reading of license plates and readings of counterfeiting and contamination by using two-dimensional image data and three-dimensional depth data | |
CN117437234B (en) | Aerial photo ground object classification and change detection method based on graph neural network |
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 |