Anti-interference smoke and fire intelligent filtering identification method based on visible light
By designing a visible light-based intelligent anti-interference pyrotechnic filtration recognition method based on visible light, using adaptive fuzzy classification model, optimized sky classification algorithm and transfer learning and integrated learning technology, the problem of poor processing of lens blur, sky cloud layer and water surface reflection in traditional methods is solved, and high-precision and stable pyrotechnic recognition is achieved.
Patent Information
- Application Number
- CN202510145283.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-10
- Publication Date
- 2025-06-17
AI Technical Summary
Traditional image processing methods are poor in dealing with interference factors such as lens blur, sky clouds and water surface reflection, which affects the accuracy and stability of firework recognition.
An intelligent anti-interference pyrotechnic filtration recognition method based on visible light was designed. By constructing a system including image acquisition, preprocessing, preliminary pyrotechnic identification, fuzzy detection, sky area identification, water surface area identification and pyrotechnic identification output modules, an adaptive fuzzy classification model, an optimized sky classification algorithm and transfer learning and integrated learning technology, the interference factors are comprehensively eliminated and high-precision pyrotechnic identification is achieved.
By systematically identifying and filtering environmental interference such as lens blur, sky clouds and water surface reflections, the stability and accuracy of pyrotechnic identification are significantly improved, and are suitable for pyrotechnic monitoring in various complex environments.
Smart Images
Figure CN120164009A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical fields of image processing and computer vision, and particularly to an anti-interference intelligent filtering and recognition method for fireworks based on visible light. Background Art
[0002] In modern fireworks monitoring and recognition, achieving high-accuracy and high-efficiency fireworks detection is crucial for public safety. Although traditional technologies have made progress in fireworks monitoring, they still face significant challenges in practical applications. Traditional image processing methods often perform poorly when dealing with interference factors such as lens blur, sky clouds, and water surface reflections. Lens blur can lead to a decrease in image quality and affect the recognition accuracy of fireworks; clouds in the sky and solar reflections on the water surface may cover up the features of fireworks or mislead the system's judgment, increasing the difficulty of recognition. Although existing image preprocessing technologies can perform basic clarification processing, their effects are limited when facing complex interferences. To improve the accuracy and stability of fireworks recognition, a new technical solution is urgently needed that can intelligently handle these environmental interference problems and thus improve the overall performance of the fireworks monitoring system.
[0003] Disadvantages of the prior art: Traditional image processing methods often perform poorly when dealing with interference factors such as lens blur, sky clouds, and water surface reflections, and their effects are limited when facing complex interferences. Summary of the Invention
[0004] An anti-interference intelligent filtering and recognition method for fireworks based on visible light provided by the present invention can comprehensively eliminate interference factors and achieve high-precision fireworks recognition.
[0005] To achieve the above object, a key aspect of an anti-interference intelligent filtering and recognition method for fireworks based on visible light provided by the present invention includes the following steps:
[0006] Step 1: Construct an anti-interference intelligent filtering and recognition system for fireworks based on visible light, where the anti-interference intelligent filtering and recognition system is provided with an image acquisition module, a preprocessing module, a preliminary fireworks recognition module, a blur detection module, a sky area recognition module, a water surface area recognition module, and a fireworks recognition output module that are connected in sequence;
[0007] Step 2: The image acquisition module real-time collects image data a of the target area and transmits it to the preprocessing module;
[0008] Step 3: The preprocessing module performs preprocessing operations on the image data a to obtain standard image data b and transmits it to the preliminary fireworks recognition module;
[0009] Step 4: The initial fireworks recognition module extracts the fireworks features from the standard image data b, initially recognizes the fireworks information in the image, obtains the fireworks image c, and transmits it to the fuzzy detection module;
[0010] Step 5: The fuzzy detection module performs fuzzy detection on the fireworks image c, excludes the fuzzy images, obtains the clear fireworks image d, and transmits it to the sky region recognition module;
[0011] Step 6: The sky region recognition module performs sky region recognition on the clear fireworks image d, excludes the sky region images, obtains the non-sky region fireworks image e, and transmits it to the water region recognition module;
[0012] Step 7: The water region recognition module performs water region recognition on the non-sky region fireworks image e, excludes the sunlight reflection interference images in the water region, obtains the target fireworks image f, and transmits it to the fireworks recognition output module;
[0013] Step 8: The fireworks recognition output module extracts the fireworks features from the target fireworks image f, recognizes and outputs the fireworks recognition result g.
[0014] The present invention performs lens blur detection on the images collected by the camera through an adaptive fuzzy classification model, and screens out the distorted images due to lens blur; uses the optimized sky classification algorithm to regard the recognized sky region as a false alarm caused by clouds and excludes it; adopts transfer learning and ensemble learning techniques to classify the images for water surface, recognizes and excludes the misjudgment caused by water surface reflection, and finally realizes the high-precision recognition of fireworks.
[0015] Through the above design, by systematically identifying and filtering environmental interferences such as lens blur, sky clouds, and water surface reflection, the stability and accuracy of fireworks recognition are significantly improved.
[0016] Preferably: In the step 2, the image acquisition module collects the image data a of the target area in real time through a high-resolution visible light camera.
[0017] Use a high-resolution visible light camera to perform real-time image acquisition, monitor the target area to obtain clear image data.
[0018] Preferably: In the step 3, the preprocessing operations include but are not limited to image denoising, color correction, and automatic white balance operations to ensure the consistency of image quality and provide a reliable data basis for subsequent processing.
[0019] Preferably: in the step 4, the preliminary fireworks recognition module first extracts the features related to fireworks in the standard image data b by using multi-scale analysis and feature fusion techniques, and then analyzes the extracted features through a fireworks recognition model to preliminarily identify the fireworks information in the image.
[0020] The preliminary fireworks recognition module uses a specially trained fireworks recognition model to analyze the extracted features and preliminarily determine whether there are fireworks in the image.
[0021] Preferably: the features related to fireworks include but are not limited to the color, brightness, and movement trajectory of fireworks.
[0022] Preferably: in the step 5, the blur detection module combines a convolutional neural network CNN with an attention mechanism to enhance the attention to the blurred areas in the image, identify the blurred parts in the image, and then mark the identified blurred images as potential false alarms and filter them out.
[0023] The preliminary fireworks recognition module combines a convolutional neural network CNN and an attention mechanism to accurately evaluate the clarity of the image. Through the adaptive attention module, it can more effectively identify and exclude blurred images, improve the accuracy of blur detection, ensure that only clear images are processed, and avoid the impact of blur on fireworks recognition.
[0024] The preliminary fireworks recognition module strengthens the attention to the blurred areas in the image through the attention mechanism and simultaneously trains with the blurred samples generated by the generative adversarial network GAN, significantly improving the recognition ability for blurred images. Only the images determined to be clear are retained and enter the sky region recognition module, thereby improving the accuracy of subsequent classification and the recognition accuracy of the system for fireworks.
[0025] Preferably: in the step 6, the sky region recognition module uses a sky classification model that combines an attention mechanism and multi-modal technology to analyze the color, texture, and shape features in the image and identify the sky and cloud regions existing in the image;
[0026] The sky region recognition module assigns a weight α i to each image feature f i of the clear fireworks image d, thereby generating the sky region S, and the expression is as follows:
[0027]
[0028] where, f i (d) is the i-th feature in the image d; α i is the attention weight related to the i-th feature, indicating the importance of this feature to the sky region.
[0029] The sky region recognition module uses an optimized classification model combined with an attention mechanism to directly identify and exclude sky regions. The sky classification model effectively reduces the interference of clouds on the recognition of fireworks by enhancing the attention to sky and cloud regions, and excludes possible cloud false alarms from candidate images.
[0030] The attention mechanism can help the model focus on the regions in the image where sky and clouds may exist, so as to improve the recognition accuracy.
[0031] To enhance the generalization ability of the sky classification model, a generative adversarial network (GAN) is introduced to generate images, thereby expanding the training dataset. Through the generator G, a latent space vector Z is input to generate fake images The expression is as follows:
[0032]
[0033] The generated fake images are adversarially trained with real images d, and the generator learns how to generate more realistic sky images. The goal of the generative adversarial network is to make the generated fake images look as much like real images as possible, so as to enhance the diversity and complexity of the training set. In this way, the generated images can help the model adapt to different sky scenes and improve the model's ability to recognize sky regions under different weather and lighting conditions.
[0034] The sky region recognition module uses an optimized sky classification algorithm to deeply analyze the image. By introducing a generative adversarial network to expand the training data and combining the attention mechanism to strengthen the recognition ability of the sky region. It can not only directly identify and exclude the sky region in the image, but also effectively handle the smoke false alarms caused by clouds. Through adversarial sample training and model parameter optimization, the system can accurately identify and exclude cloud interference under various weather conditions, thereby further improving the accuracy of fireworks recognition.
[0035] Preferably: the attention weight α i is automatically adjusted through training, so that the model can better focus on the sky region. The attention weight α i is calculated based on the similarity of image features, and the expression is as follows:
[0036]
[0037] where w i is the learned weight vector that controls the importance of each feature; the superscript T represents the vector transpose.
[0038] Preferably: in the step 7, the water surface area recognition module uses a model based on dynamic threshold adjustment and deep learning to recognize the water surface area through color space conversion and reflection feature detection, and combines an attention mechanism to focus on the areas affected by reflection;
[0039] The water surface area recognition module first uses a pre-trained deep learning model M pre to extract the feature F of the non-sky area fireworks image e, and the expression is as follows:
[0040] F = M pre (e, θ pre )
[0041] where F is the feature representation of the non-sky area fireworks image e, and θ pre is the parameter of the pre-trained deep learning model M pre ;
[0042] The water surface area recognition module fine-tunes the parameters of the pre-trained deep learning model M according to the extracted feature F to obtain the fine-tuned deep learning model M pre , and then uses the fine-tuned deep learning model M fine to recognize the water surface area W in the non-sky area fireworks image e, and the expression is as follows: fine W = M
[0043] (F, θ fine ) fine )
[0044] where θ fine is the parameter of the fine-tuned deep learning model M fine . The fine-tuned model is optimized specifically for the water surface area to more accurately recognize the water surface area in the image, especially the areas with greater reflection influence.
[0045] The water surface area recognition module applies color analysis and reflected light feature detection technologies, combines dynamic threshold adjustment and deep learning models to recognize the water surface area in the image and remove the reflected sunlight spots. The classification algorithm optimized by reinforcement learning effectively excludes the reflection interference of the water surface area, thereby improving the accuracy of fireworks recognition.
[0046] The water surface area recognition module adopts transfer learning and ensemble learning technologies to detect water surface interference by analyzing the water surface reflected light features. The water surface area recognition module combines deep learning models and generates adversarial samples for training to improve the recognition ability of water surface reflections. The ensemble learning technology further improves the stability and accuracy of the model by fusing the outputs of multiple classifiers, effectively avoiding misidentifying water surface reflections as fireworks, thereby ensuring high accuracy and reliability of the system in complex environments.
[0047] Preferably, in step 8, the fireworks recognition output module outputs the fireworks recognition result g to the control system, triggering the corresponding recording and alarm mechanisms.
[0048] The image after blurring, sky cloud, and water surface reflection filtering will undergo final verification. If the image passes all interference elimination steps and conforms to the fireworks characteristics, it is confirmed as a real fireworks event.
[0049] The fireworks recognition output module outputs the recognition result and triggers the corresponding monitoring and response system to achieve real-time monitoring and rapid response to fireworks.
[0050] Advantages of the present invention: The present invention provides an efficient and reliable fireworks recognition solution, effectively solving the environmental interference problem in traditional methods, greatly improving the accuracy and stability of fireworks detection, and being applicable to fireworks monitoring in various complex environments;
[0051] This method comprehensively eliminates interference factors through lens blur classification, sky classification, and water surface classification to achieve high-precision fireworks recognition. Description of the Drawings
[0052] Figure 1 It is a schematic flow chart of the present invention. Detailed Embodiment
[0053] The present invention will be further described in detail below with reference to the drawings and specific examples. The following examples or drawings are used to illustrate the present invention but not to limit the scope of the present invention.
[0054] As Figure 1 shown: An anti-interference fireworks intelligent filtering and recognition method based on visible light includes the following steps:
[0055] Step 1: Construct an anti-interference fireworks intelligent filtering and recognition system based on visible light. The anti-interference fireworks intelligent filtering and recognition system is provided with an image acquisition module, a preprocessing module, a preliminary fireworks recognition module, a blur detection module, a sky area recognition module, a water surface area recognition module, and a fireworks recognition output module connected in sequence;
[0056] Step 2: The image acquisition module real-time collects the image data a of the target area and transmits it to the preprocessing module;
[0057] Step 3: The preprocessing module performs preprocessing operations on the image data a to obtain standard image data b and transmits it to the preliminary fireworks recognition module;
[0058] Step 4: The preliminary fireworks recognition module extracts the fireworks characteristics from the standard image data b, preliminarily recognizes the fireworks information in the image to obtain a fireworks image c, and transmits it to the blur detection module;
[0059] Step 5: The blur detection module performs blur detection on the fireworks image c, excludes blurred images, obtains a clear fireworks image d, and transmits it to the sky area recognition module;
[0060] Step 6: The sky area recognition module performs sky area recognition on the clear fireworks image d, excludes sky area images, obtains a non-sky area fireworks image e, and transmits it to the water area recognition module;
[0061] Step 7: The water area recognition module performs water area recognition on the non-sky area fireworks image e, excludes the sunlight reflection interference image in the water area, obtains a target fireworks image f, and transmits it to the fireworks recognition output module;
[0062] Step 8: The fireworks recognition output module extracts fireworks features from the target fireworks image f, identifies and outputs the fireworks recognition result g.
[0063] In step 2, the image acquisition module collects the image data a of the target area in real time through a high-resolution visible light camera.
[0064] In step 3, the preprocessing operations include but are not limited to image denoising, color correction, and automatic white balance operations.
[0065] In step 4, the fireworks preliminary recognition module first extracts the features related to fireworks in the standard image data b by using multi-scale analysis and feature fusion techniques, and then analyzes the extracted features through a fireworks recognition model to preliminarily identify the fireworks information in the image.
[0066] The features related to fireworks include but are not limited to the color, brightness, and motion trajectory of fireworks.
[0067] In step 5, the blur detection module combines the convolutional neural network CNN with the attention mechanism to enhance the attention to the blurred area of the image, identifies the blurred part in the image, and then marks the identified blurred image as a potential false alarm and filters it out.
[0068] In step 6, the sky area recognition module uses a sky classification model that combines the attention mechanism and multi-modal technology to analyze the color, texture, and shape features in the image, and identifies the sky and cloud areas existing in the image;
[0069] The sky area recognition module assigns a weight α i to each image feature f i of the clear fireworks image d through the attention mechanism, so as to generate the sky area S, and the expression is as follows:
[0070]
[0071] Among them, f i (d) is the i-th feature in image d; α i is the attention weight related to the i-th feature, indicating the importance of this feature to the sky region.
[0072] The attention weight α i is automatically adjusted through training, enabling the model to better focus on the sky region. The calculation method of the attention weight α i is based on the similarity of image features, and the expression is as follows:
[0073]
[0074] Among them, w i is the learned weight vector, controlling the importance of each feature; the superscript T represents vector transpose.
[0075] In the above-mentioned step 7, the water surface area recognition module uses a model based on dynamic threshold adjustment and deep learning to recognize the water surface area through color space conversion and reflection feature detection, and combines the attention mechanism to focus on the areas affected by reflection;
[0076] The water surface area recognition module first uses the pre-trained deep learning model M pre to extract the features F of the non-sky region fireworks image e, and the expression is as follows:
[0077] F = M pre (e, θ pre )
[0078] Among them, F is the feature representation of the non-sky region fireworks image e, and θ pre is the parameter of the pre-trained deep learning model M pre ;
[0079] The water surface area recognition module fine-tunes the parameters of the pre-trained deep learning model M pre according to the extracted features F to obtain the fine-tuned deep learning model M fine , and then uses the fine-tuned deep learning model M fine to recognize the water surface area W in the non-sky region fireworks image e, and the expression is as follows:
[0080] W = M fine (F, θ fine )
[0081] Among them, θ fine is the fine-tuned deep learning model M fineParameters. The fine-tuned model is specifically optimized for water surface areas to more accurately identify water surface areas in images, especially areas with significant reflection effects.
[0082] In step 8, the fireworks recognition output module outputs the fireworks recognition result g to the control system, triggering corresponding recording and alarm mechanisms.
[0083] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.
Claims
1. A visible light-based anti-interference fireworks intelligent filtering and identification method, characterized in that: The following steps are involved: Step 1: construct an anti-interference fireworks intelligent filtering and recognition system based on visible light, wherein the anti-interference fireworks intelligent filtering and recognition system is provided with an image acquisition module, a preprocessing module, a fireworks preliminary recognition module, a fuzzy detection module, a sky area recognition module, a water surface area recognition module and a fireworks recognition output module connected in sequence; Step 2: The image acquisition module acquires image data a of the target area in real time and transmits it to the preprocessing module; Step 3: the preprocessing module performs preprocessing operations on the image data a to obtain standard image data b, and transmits the standard image data b to the fireworks preliminary recognition module; Step 4: The fireworks preliminary recognition module extracts fireworks features from the standard image data b, preliminarily recognizes fireworks information in the image, obtains fireworks image c, and transmits it to the blur detection module; Step 5: The blur detection module performs blur detection on the fireworks image c, removes the blurry image, obtains a clear fireworks image d, and transmits it to the sky area recognition module; Step 6: The sky area recognition module performs sky area recognition on the clear fireworks image d, excludes the sky area image, obtains the non-sky area fireworks image e, and transmits it to the water surface area recognition module; Step 7: The water surface area recognition module performs water surface area recognition on the non-sky area fireworks image e, eliminates the interference image of sunlight reflection in the water surface area, obtains the target fireworks image f, and transmits it to the fireworks recognition output module; Step 8: The fireworks recognition output module extracts fireworks features from the target fireworks image f, recognizes and outputs fireworks recognition result g.
2. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In the step 2, the image acquisition module acquires image data a of the target area in real time through a high-resolution visible light camera.
3. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In step 3, the preprocessing operation includes but is not limited to image denoising, color correction and automatic white balance operations.
4. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In step 4, the fireworks preliminary recognition module first uses multi-scale analysis and feature fusion technology to extract features related to fireworks in the standard image data b, and then analyzes the extracted features through a fireworks recognition model to preliminarily recognize the fireworks information in the image.
5. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 4 is characterized in that: Features related to fireworks include, but are not limited to, color, brightness, and movement trajectory of fireworks.
6. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In step 5, the blur detection module enhances attention to the blurred area of the image through the convolutional neural network CNN combined with the attention mechanism, identifies the blurred part in the image, and then marks the identified blurred image as a potential false alarm and filters it out.
7. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In step 6, the sky area recognition module uses a sky classification model that integrates attention mechanism and multimodal technology to analyze the color, texture and shape features in the image and identify the sky and cloud areas in the image; The sky area recognition module uses an attention mechanism to identify each image feature f of the clear fireworks image d. i Assign a weight α i , thus generating the sky area S, the expression is as follows: Among them, f i (d) is the i-th feature in image d; α i is the attention weight associated with the i-th feature, indicating the importance of this feature to the sky area.
8. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 7 is characterized in that: The attention weight α i The attention weight α is automatically adjusted through training i The calculation method of is based on the similarity of image features, and the expression is as follows: Among them, w i is the learned weight vector that controls the importance of each feature; the function exp(·) is the natural exponential function, and the superscript T represents the vector transpose.
9. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In step 7, the water surface area recognition module first uses the pre-trained deep learning model M pre The feature F of the fireworks image e in the non-sky area is extracted, and the expression is as follows: F=M pre (e,θ pre ) Among them, F is the feature representation of the fireworks image e in the non-sky area, θ pre is the pre-trained deep learning model M pre parameter; The water surface area recognition module uses the pre-trained deep learning model M according to the extracted features F. pre The parameters are fine-tuned to obtain the fine-tuned deep learning model M fine , and then use the fine-tuned deep learning model M fine Identify the water surface area W in the non-sky area fireworks image e, the expression is as follows: W=M fine (F,θ fine ) Among them, θ fine is the fine-tuned deep learning model M fine parameter.
10. The method for intelligent filtering and identifying fireworks based on visible light with anti-interference according to claim 1 is characterized in that: In step 8, the fireworks recognition output module outputs the fireworks recognition result g to the control system, triggering a corresponding recording and alarm mechanism.