Image processing method and system based on AI recognition
Feature extraction through multi-spectral sensor acquisition and sparse representation algorithm, combined with adaptive detection model, the problem of insufficient accuracy and real-time target recognition in dynamically changing scenarios in the prior art is solved, and more efficient image processing and decision-making management are achieved.
Patent Information
- Application Number
- CN202410557091.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-05-07
AI Technical Summary
Existing AI-based image processing methods are insufficient in accuracy and lack of real-time performance when processing complex and dynamically changing scenes. Especially when the target object and background characteristics are similar or the target object are constantly changing, it is difficult to accurately extract the target object. Features and lack adaptive adjustment capabilities.
The AI imaging device with built-in multi-spectral sensors collects spectral images of dynamically changing scenes, pre-processes them through preset physical models, uses sparse representation algorithms to perform multi-level feature extraction, and inputs features into the trained adaptive detection model for Real-time detection and decision-making management.
It improves the accuracy of target recognition and real-time monitoring capabilities, can more accurately capture the detailed characteristics of target objects in dynamically changing scenarios, enhances the system's adaptability and processing capabilities to complex environments, and realizes efficient decision-making management.
Smart Images

Figure CN118279748B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to an image processing method and system based on AI recognition. Background Art
[0002] With the rapid development of artificial intelligence and image processing technology, AI-based image recognition technology has been widely used in agricultural monitoring, environmental monitoring, urban planning and other fields. These technologies identify and classify target objects in images by analyzing various features in images, such as shape, color, texture, etc. In agricultural monitoring, the growth status and pests of crops can be monitored by analyzing images; in environmental monitoring, the degree of pollution can be assessed by analyzing images of water bodies and soil; in urban planning, urban transportation can be optimized by analyzing images of roads and traffic.
[0003] In the prior art, some AI-based image processing methods have been proposed. These methods usually include steps such as image acquisition, preprocessing, feature extraction, and target recognition. In the image acquisition stage, high-precision cameras or sensors are used to collect images of the target scene; in the preprocessing stage, image quality is improved by means such as noise reduction and contrast enhancement; in the feature extraction stage, various algorithms are used to extract the key features of the image; in the target recognition stage, the extracted features are input into the trained model to classify or detect the target. However, these prior arts often face challenges of insufficient accuracy and poor real-time performance when dealing with complex scenes and dynamically changing targets.
[0004] The main technical defects of the prior art are its insufficient accuracy and real-time performance in target recognition in dynamically changing scenes. First, traditional feature extraction methods have difficulty in accurately extracting the key features of the target object when dealing with complex dynamic scenes, especially when the target object and the background features are similar or the target object is constantly changing. Secondly, although the existing AI models perform well in the recognition of static images, they often cannot meet the requirements of efficiency and real-time when processing real-time dynamically changing image sequences. In addition, the prior art generally lacks the ability to adaptively adjust dynamically changing targets in decision management, which limits its effectiveness and reliability in actual application scenarios. Therefore, there is an urgent need for an AI image processing method that can efficiently and accurately handle target recognition in dynamically changing scenes. Summary of the invention
[0005] The present invention provides an image processing method and system based on AI recognition, which are used to solve the above-mentioned technical problems.
[0006] A first aspect of the present invention provides an image processing method based on AI recognition, the image processing method based on AI recognition comprising:
[0007] The images of dynamically changing scenes are collected by an AI imaging device with a built-in multispectral sensor to obtain spectral images of dynamically changing scenes;
[0008] Inputting the spectral image into a preset physical model for image preprocessing to obtain a spectral image after image preprocessing;
[0009] Based on a preset sparse representation algorithm, multi-level feature extraction is performed on the spectral image after image preprocessing to obtain multi-level features of the target object; wherein the target object includes at least crops, water bodies, and road traffic; the multi-level features of the target object include at least contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features, and surrounding environment background features of the target object;
[0010] Inputting the multi-level features of the target object into the trained adaptive detection model for real-time detection to obtain the multi-level feature detection results of the target object; wherein the adaptive detection model is obtained through pre-training;
[0011] The multi-level feature detection results are analyzed to obtain analysis results, and decision management is performed on the target object based on the analysis results.
[0012] Optionally, in a first implementation of the first aspect of the present invention, the AI imaging device with a built-in multispectral sensor collects images of dynamically changing scenes to obtain spectral images of the dynamically changing scenes, including:
[0013] Automatically adjust the spectral response range of AI imaging devices with built-in multispectral sensors under different ambient lighting conditions to obtain optimized spectral image acquisition parameters;
[0014] Based on the optimized spectral image acquisition parameters, the image of the dynamically changing scene is preliminarily acquired by an AI imaging device with a built-in multispectral sensor to obtain spectral features within a specific wavelength range;
[0015] Applying a preset spectral analysis algorithm to extract spectral features within a specific wavelength range, identifying spectral features in each frame, and obtaining multi-frame spectral features; wherein the spectral features in each frame are used to represent material composition and state changes in a dynamically changing scene;
[0016] The spectral features of multiple frames collected continuously are integrated to construct a spectral image of a dynamically changing scene; wherein the spectral image of the dynamically changing scene is used to represent the change trajectory of the spectral information of the target object in the dynamically changing scene over time.
[0017] Optionally, in a second implementation of the first aspect of the present invention, the step of inputting the spectral image into a preset physical model for image preprocessing to obtain the spectral image after image preprocessing includes:
[0018] Obtain a preset physical model, and input the spectral image into the preset physical model for image preprocessing; wherein the preset physical model at least includes an optical distortion correction sub-model, a color deviation adjustment sub-model, a noise suppression sub-model, and a dynamic range adjustment sub-model;
[0019] Analyzing the radial distortion and tangential distortion of the spectral image by using the optical distortion correction sub-model, calculating a first correction parameter corresponding to the radial distortion and a second correction parameter corresponding to the tangential distortion, and adjusting the position of each pixel in the spectral image according to the first correction parameter and the second correction parameter;
[0020] Adjusting the color balance of the spectral image and compensating for the color deviation of the spectral image by using the color deviation adjustment sub-model;
[0021] Removing random noise and color noise in the spectral image by using the noise suppression sub-model;
[0022] The dynamic range of the spectral image is adjusted by the dynamic range adjustment sub-model to balance the bright and dark parts of the image;
[0023] The spectral image is processed by the optical distortion correction sub-model, the color deviation adjustment sub-model, the noise suppression sub-model and the dynamic range adjustment sub-model respectively to obtain a spectral image after image preprocessing.
[0024] Optionally, in a third implementation of the first aspect of the present invention, the multi-level feature extraction is performed on the spectral image after image preprocessing based on a preset sparse representation algorithm to obtain the multi-level features of the target object, including:
[0025] Based on the preset spectral image segmentation rules, the preprocessed spectral image is segmented to obtain corresponding image blocks;
[0026] Based on a preset sparse representation algorithm, sparse encoding is performed on each image block to obtain a sparse representation result of each image block;
[0027] Based on the sparse representation result of each image block, hierarchical feature extraction is performed on the preprocessed spectral image to obtain multi-level features of the target object.
[0028] Optionally, in a fourth implementation of the first aspect of the present invention, the training process of the adaptive detection model includes:
[0029] Collect training data and input the training data into a preliminarily constructed adaptive deep learning network; wherein the adaptive deep learning network includes at least a shape feature analysis sub-model, a texture analysis sub-model, a color recognition sub-model, a motion tracking sub-model, and a background separation sub-model; and the training data is a multi-level feature of the target object;
[0030] Based on the shape feature analysis sub-model, the contour features and shape features of the target object in the spectral image are analyzed to identify the basic structure of the target object and output the shape feature index;
[0031] Based on the texture analysis sub-model, the texture features of the surface of the target object in the spectral image are analyzed and a texture feature index is output; wherein the texture features at least include the roughness, directionality and repetitive pattern of the texture;
[0032] Based on the color recognition sub-model, the color characteristics and color distribution characteristics of the target object in the spectral image are identified and analyzed, the contribution of color to the recognition of the target object is evaluated, and the color characteristic index is output;
[0033] Based on the motion tracking sub-model, the dynamic image sequence characteristics of the target object in the spectral image are analyzed, the motion trajectory and speed of the target object are tracked, the motion pattern is analyzed, and the motion characteristic index is output;
[0034] Based on the background separation sub-model, analyze the surrounding environment background features of the target object in the spectral image, separate the target object from the surrounding environment background features, and output a background separation effect index;
[0035] Extracting a preset actual index value from the collected image data; wherein the actual index value represents the characteristic performance of the target object under an ideal state and is used as a reference standard in the training process;
[0036] Calculate an error value based on the deviation between the predicted index output by each sub-model in the constructed adaptive deep learning network and the preset actual index value; wherein the error value reflects the difference between the predicted index and the actual index value;
[0037] Based on the preset adaptive learning rate adjustment strategy, the parameters of each sub-model in the adaptive deep learning network are gradually adjusted to minimize the error value until the training of the adaptive deep learning network is completed to obtain the trained adaptive detection model.
[0038] Optionally, in the fifth implementation of the first aspect of the present invention, the actual index value includes at least actual contour features, actual shape features, actual texture features, actual color features, actual color distribution features, actual dynamic image sequence features, and actual surrounding environment background features.
[0039] A second aspect of the present invention provides an image processing system based on AI recognition, the image processing system based on AI recognition comprising:
[0040] An acquisition module is used to acquire images of dynamically changing scenes through an AI imaging device with a built-in multispectral sensor to obtain spectral images of dynamically changing scenes;
[0041] An image preprocessing module, used for inputting the spectral image into a preset physical model for image preprocessing to obtain a spectral image after image preprocessing;
[0042] An extraction module is used to perform multi-level feature extraction on the spectral image after image preprocessing based on a preset sparse representation algorithm to obtain multi-level features of the target object; wherein the target object includes at least crops, water bodies, and road traffic; the multi-level features of the target object include at least contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features, and surrounding environment background features of the target object;
[0043] A detection module, used for inputting the multi-level features of the target object into the trained adaptive detection model for real-time detection to obtain the multi-level feature detection results of the target object; wherein the adaptive detection model is obtained through pre-training;
[0044] The management module is used to analyze the multi-level feature detection results, obtain analysis results, and make decision management on the target object based on the analysis results.
[0045] The third aspect of the present invention provides an image processing device based on AI recognition, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the image processing device based on AI recognition executes the above-mentioned image processing method based on AI recognition.
[0046] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, which, when executed on a computer, enable the computer to execute the above-mentioned image processing method based on AI recognition.
[0047] The technical solution provided by the present invention has the following beneficial effects: the present invention provides an image processing method and system for AI recognition, wherein an AI imaging device with a built-in multispectral sensor collects images of dynamically changing scenes to obtain spectral images of dynamically changing scenes; the spectral images are input into a preset physical model for image preprocessing to obtain spectral images after image preprocessing; based on a preset sparse representation algorithm, multi-level feature extraction is performed on the spectral images after image preprocessing to obtain multi-level features of the target object; the multi-level features of the target object are input into a trained adaptive detection model for real-time detection to obtain multi-level feature detection results of the target object; the multi-level feature detection results are analyzed to obtain analysis results, and decision management is performed on the target object based on the analysis results. The present invention uses spectral images collected by multispectral sensors and combines them with physical models for preprocessing, and applies sparse representation algorithms for in-depth multi-level feature extraction. This method can more accurately capture the detailed features of target objects in dynamically changing scenes. This not only improves the accuracy of target recognition, but also ensures rapid response in a dynamically changing environment, significantly improving the ability of real-time monitoring and recognition. The present invention utilizes the multispectral image acquisition capability of AI imaging equipment combined with advanced image processing technology to effectively handle a variety of complex scenes including crop growth, water monitoring, road traffic conditions, etc. By accurately extracting and analyzing the multi-level features of target objects in different scenes, the system's ability to adapt to and handle complex environments is enhanced. By inputting the extracted multi-level features into a well-trained adaptive detection model, the present invention can not only detect the target object in real time, but also make efficient decision-making management based on the analysis results of the deep learning model. This adaptive detection and decision-making mechanism enables the system to make more accurate and reasonable management decisions for different targets and scenes. The present invention adopts AI technology and deep learning models. The method can automatically identify and analyze a large amount of complex image data, reduce the need for manual intervention, and significantly improve the intelligent level of data processing. This intelligent processing not only improves efficiency, but also enhances the system's ability to learn and adapt to new situations. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 A schematic diagram of an embodiment of an image processing method based on AI recognition in an embodiment of the present invention;
[0049] Figure 2 Schematic diagram of an embodiment of an image processing system based on AI recognition in an embodiment of the present invention. DETAILED DESCRIPTION
[0050] Embodiments of the present invention provide an image processing method and system based on AI recognition. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0051] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the image processing method based on AI recognition in an embodiment of the present invention includes:
[0052] Step 101: collect images of dynamically changing scenes through an AI imaging device with a built-in multispectral sensor to obtain a spectral image of the dynamically changing scene;
[0053] It is understandable that the execution subject of the present invention may be an image processing system based on AI recognition, or may be a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.
[0054] Specifically, step 101: Multispectral sensor AI imaging device dynamic scene image acquisition
[0055] Dynamic scene spectral image acquisition:
[0056] AI imaging devices with built-in multispectral sensors are used to capture images of dynamically changing scenes. Multispectral sensors can capture multi-band spectral information and improve the information acquisition capabilities of images.
[0057] In dynamic scenes, the fast response and high frame rate of the device can achieve continuous acquisition of spectral images under scene changes, ensuring the continuity and integrity of the image.
[0058] Spectral image processing:
[0059] The collected spectral images need to be preprocessed and analyzed, including denoising, registration, band fusion and other processing steps to improve image quality and information accuracy.
[0060] Apply AI algorithms for intelligent processing of spectral images, such as spectral feature extraction and analysis based on deep learning, to achieve efficient recognition and analysis of dynamically changing scenes.
[0061] Example:
[0062] AI imaging devices equipped with multispectral sensors, such as the Hyperspectral Imaging (HSI) system, are used to collect spectral images of crops in the agricultural field. The equipment can obtain crop growth information in different bands and realize dynamic monitoring of crop health.
[0063] The collected spectral images are quickly preprocessed and analyzed by AI image processing algorithms to accurately identify the growth status and pest and disease conditions of crop leaves.
[0064] Step 102: input the spectral image into a preset physical model for image preprocessing to obtain a spectral image after image preprocessing;
[0065] Specifically, step 102: spectral image preprocessing
[0066] Input spectral image to physical model:
[0067] The collected spectral image is input into a preset physical model, which is designed based on optical principles and spectral characteristics and is used for image preprocessing and correction.
[0068] The physical model can remove noise, adjust brightness and contrast, and perform color correction on spectral images to improve image quality and accuracy.
[0069] Image preprocessing:
[0070] The physical model preprocesses the spectral image, including correcting the geometric deformation of the image, wavelength correction, removing the influence of atmosphere or water vapor, etc., to obtain more accurate spectral information.
[0071] By preprocessing the spectral image, interference factors can be eliminated, errors can be reduced, and the image analysis and recognition capabilities can be improved.
[0072] Example:
[0073] The spectral image of the dynamic scene is input into the HysiCAD (Hyperspectral Imaging Computer Aided Design) physical model for processing. The HysiCAD model is based on the principles of optics and spectroscopy and can perform geometric correction and wavelength correction on the spectral image.
[0074] Spectral images preprocessed by HysiCAD have higher clarity and accuracy, and are suitable for precise spectral analysis and diagnosis in fields such as agriculture and medical imaging.
[0075] Step 103: Based on a preset sparse representation algorithm, multi-level feature extraction is performed on the spectral image after image preprocessing to obtain multi-level features of the target object; wherein the target object includes at least crops, water bodies, and road traffic; the multi-level features of the target object include at least contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features, and surrounding environment background features of the target object;
[0076] Specifically, step 103: multi-level feature extraction
[0077] Sparse representation algorithm application:
[0078] Based on the preset sparse representation algorithm, multi-level feature extraction is performed on the spectral image after image preprocessing. The sparse representation algorithm is a feature extraction method based on the principle of data sparsity, which can effectively capture the characteristic information of the target object.
[0079] Using the sparse representation algorithm, multi-level features of target objects such as crops, water bodies, road traffic, etc. are extracted from spectral images, including contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features and surrounding environment background features.
[0080] Multi-level target object feature extraction:
[0081] For target objects such as crops, sparse representation algorithms can extract their specific farmland contour features, crop shape features, and crop texture features to facilitate crop type identification and growth status monitoring.
[0082] For water targets, the algorithm can extract the color characteristics and color distribution characteristics of the water body for water quality monitoring and environmental pollution monitoring.
[0083] The feature extraction of road traffic targets can include the vehicle's motion trajectory features, vehicle color features and surrounding road background features, which are used for traffic flow analysis and vehicle identification.
[0084] Example:
[0085] By using the SIFT (Scale-Invariant Feature Transform) method based on the sparse representation algorithm, feature extraction is performed on the spectral image, and multi-level feature extraction and classification recognition of different target objects are achieved.
[0086] In the agricultural field, sparse representation algorithms are used to extract the characteristics of various crops in the farmland, such as the outline of cereal crops and the shape of fruit trees, providing more accurate farmland management and crop monitoring methods for smart agriculture.
[0087] Step 104: input the multi-level features of the target object into the trained adaptive detection model for real-time detection to obtain a multi-level feature detection result of the target object; wherein the adaptive detection model is obtained through pre-training;
[0088] Specifically, step 104: real-time multi-level feature detection of target objects
[0089] Multi-level features are input to the adaptive detection model:
[0090] The multi-level features of the target object extracted by the sparse representation algorithm are input into the pre-trained adaptive detection model for real-time detection. The adaptive detection model is built based on deep learning and neural network technology, which can efficiently identify and detect target objects.
[0091] The adaptive detection model achieves accurate detection of target objects by learning the relationship between multi-level features, including the contour, shape, color and other features of the target object.
[0092] Multi-level feature detection result output:
[0093] The adaptive detection model detects the input multi-level features in real time and outputs the multi-level feature detection results of the target object. The detection results include detailed information such as the location, category, and attributes of the target object, providing basic data for further analysis and processing.
[0094] Combining the detection results of multi-level features, comprehensive recognition and positioning of different target objects can be achieved, providing accurate target detection and analysis for applications in various fields.
[0095] Example:
[0096] The multi-level features of different crops in the farmland are input into the YOLO (You Only Look Once) detection model trained by deep learning for real-time detection. The model can simultaneously detect multiple target objects, such as wheat, rice and other crops, and output their location and category information.
[0097] In an urban road environment, the multi-level features of road traffic targets are input into the Faster R-CNN (Faster Region-based Convolutional Neural Network) detection model for real-time detection. The model can accurately detect the location and type of vehicles, providing support for traffic management and intelligent transportation systems.
[0098] Step 105: Analyze the multi-level feature detection results to obtain analysis results, and make decision management on the target object based on the analysis results.
[0099] Specifically, step 105: multi-level feature detection result analysis and decision management
[0100] Analysis of multi-level feature detection results:
[0101] The multi-level feature detection results obtained by the adaptive detection model are deeply analyzed, including the location, category, attributes and other information of the target object. By analyzing these features, more detailed and comprehensive feature information of the target object can be obtained.
[0102] Computer vision and image processing technology are used to comprehensively analyze the multi-level feature detection results, identify and describe the various features of the target object, and provide data support for subsequent decision-making management.
[0103] Decision Management:
[0104] Based on the analysis results, formulate decision-making management strategies for target objects. According to the multi-level feature detection results, the target objects are classified, identified, and located, thus providing a scientific and reasonable basis for decision-making in actual work.
[0105] Combined with the analysis results, decision-making and management tasks such as monitoring the growth of farmland crops, detecting and controlling water pollution, and managing road traffic flow can be carried out.
[0106] Example:
[0107] In smart agriculture, by analyzing the multi-level feature detection results of crops in the farmland, we can obtain information such as crop growth status, pests and diseases, etc. Based on these analysis results, farmers can take corresponding fertilization, irrigation and prevention measures to achieve precision agricultural management.
[0108] In urban traffic management, by analyzing the multi-level characteristics of road traffic targets, it is possible to monitor vehicle driving conditions, road congestion conditions, etc. in real time. Based on the analysis results, the traffic management department can optimize the setting of traffic lights, adjust the distribution of traffic flow, etc., to improve traffic efficiency.
[0109] Another embodiment of the image processing method based on AI recognition in the embodiment of the present invention includes:
[0110] The method of collecting images of dynamically changing scenes by using an AI imaging device with a built-in multispectral sensor to obtain a spectral image of the dynamically changing scene includes:
[0111] Automatically adjust the spectral response range of AI imaging devices with built-in multispectral sensors under different ambient lighting conditions to obtain optimized spectral image acquisition parameters;
[0112] Based on the optimized spectral image acquisition parameters, the image of the dynamically changing scene is preliminarily acquired by an AI imaging device with a built-in multispectral sensor to obtain spectral features within a specific wavelength range;
[0113] Applying a preset spectral analysis algorithm to extract spectral features within a specific wavelength range, identifying spectral features in each frame, and obtaining multi-frame spectral features; wherein the spectral features in each frame are used to represent material composition and state changes in a dynamically changing scene;
[0114] The spectral features of multiple frames collected continuously are integrated to construct a spectral image of a dynamically changing scene; wherein the spectral image of the dynamically changing scene is used to represent the change trajectory of the spectral information of the target object in the dynamically changing scene over time.
[0115] Specific, important term explanation:
[0116] Multispectral sensor: A sensor that can simultaneously capture spectral information in multiple different wavelength ranges. It is widely used in remote sensing, medical imaging, agriculture and other fields.
[0117] Spectral response range: refers to the spectral range that the sensor can respond to, that is, the lowest and highest wavelengths that the sensor can detect.
[0118] Spectral feature extraction: The process of identifying and extracting useful information from spectral data to analyze the composition, state and other characteristics of matter.
[0119] Dynamically changing scene: a scene that changes continuously over time, including but not limited to changes in the natural environment, biological growth and development, chemical reaction processes, etc.
[0120] Application scenarios:
[0121] This technical solution can be applied to many fields such as environmental monitoring, precision agriculture, medical diagnosis, chemical analysis, etc. Especially in scenarios where real-time monitoring and analysis of material composition and state changes are required, such as monitoring vegetation growth, changes in water pollutants, and skin lesion detection.
[0122] Furthermore, considering an application scenario of precision agriculture, agricultural researchers need to monitor and analyze the dynamic changes of crop growth environment and crop growth status:
[0123] Spectral response range adjustment:
[0124] Based on the weather conditions of the day and the needs of the crop growth cycle, AI imaging equipment automatically adjusts the response range of the multispectral sensor to capture the spectral information required about soil moisture and crop growth health.
[0125] Preliminary spectral image acquisition:
[0126] Based on the adjusted optimal spectral image acquisition parameters, the AI imaging device captures spectral images of crops and their growth environment, with special attention paid to specific wavelength ranges that have a greater impact on crop growth conditions.
[0127] Spectral feature extraction:
[0128] Advanced spectral analysis algorithms are used to process the collected images and identify key spectral features in each frame, such as reflectance changes at specific wavelengths, to extract information about soil moisture and crop chlorophyll content.
[0129] Construct a spectral image:
[0130] The continuously collected multi-frame spectral features are integrated to form a dynamically changing spectral image, which records in detail the changing trajectory of the spectral information of the crop growth status and environmental conditions during the monitoring period.
[0131] Beneficial effects:
[0132] Improve monitoring accuracy: By fine-tuning the spectral response range and collecting key spectral features in a targeted manner, dynamically changing scenes can be monitored and analyzed more accurately.
[0133] Enhanced analysis capabilities: The use of advanced spectral analysis algorithms can extract more valuable information from spectral images, providing strong data support for scientific research.
[0134] Optimize decision support: By constructing dynamically changing spectral images, researchers can more intuitively observe every changing detail in the crop growth process and provide scientific guidance for agricultural production.
[0135] Another embodiment of the image processing method based on AI recognition in the embodiment of the present invention includes: inputting the spectral image into a preset physical model for image preprocessing to obtain the spectral image after image preprocessing, including:
[0136] Obtain a preset physical model, and input the spectral image into the preset physical model for image preprocessing; wherein the preset physical model at least includes an optical distortion correction sub-model, a color deviation adjustment sub-model, a noise suppression sub-model, and a dynamic range adjustment sub-model;
[0137] Analyzing the radial distortion and tangential distortion of the spectral image by using the optical distortion correction sub-model, calculating a first correction parameter corresponding to the radial distortion and a second correction parameter corresponding to the tangential distortion, and adjusting the position of each pixel in the spectral image according to the first correction parameter and the second correction parameter;
[0138] Adjusting the color balance of the spectral image and compensating for the color deviation of the spectral image by using the color deviation adjustment sub-model;
[0139] Removing random noise and color noise in the spectral image by using the noise suppression sub-model;
[0140] The dynamic range of the spectral image is adjusted by the dynamic range adjustment sub-model to balance the bright and dark parts of the image;
[0141] The spectral image is processed by the optical distortion correction sub-model, the color deviation adjustment sub-model, the noise suppression sub-model and the dynamic range adjustment sub-model respectively to obtain a spectral image after image preprocessing.
[0142] Specific, important term explanation:
[0143] Optical distortion correction submodel: used to correct the distortion in the image caused by the optical properties of the lens, including radial distortion and tangential distortion, to restore the correct geometric shape of the image.
[0144] Color deviation adjustment sub-model: adjusts the color balance of the image, compensates for color deviation caused by factors such as sensor characteristics and changes in light source color temperature, and ensures the authenticity and consistency of image color.
[0145] Noise suppression sub-model: Removes random noise (such as high ISO noise) and color noise (such as caused by sensor heating or electronic interference) from the image, improving image quality.
[0146] Dynamic range adjustment sub-model: Adjusts the bright and dark parts of the image, expanding the dynamic range of the image so that details can be preserved under a wide range of brightness conditions.
[0147] Application scenarios:
[0148] This technical solution can be applied to multiple fields such as remote sensing imaging, medical imaging, precision agriculture, environmental monitoring, etc. In these scenarios, the quality of spectral images directly affects the accuracy and reliability of subsequent analysis, so efficient image preprocessing is required.
[0149] Furthermore, suppose that in the field of environmental monitoring, researchers use spectral images collected by multispectral sensors to analyze vegetation coverage and water pollution levels in a certain area:
[0150] Optical distortion correction:
[0151] First, based on the collected spectral image, the radial distortion and tangential distortion parameters of the image are analyzed and calculated using the optical distortion correction submodel. Then, the position of each pixel in the image is adjusted according to these parameters to correct the image distortion.
[0152] Color bias adjustment:
[0153] Next, the color balance of the image is finely adjusted through the color deviation adjustment sub-model to compensate for the color deviation caused by changes in sensor characteristics or lighting conditions, ensuring the accuracy and consistency of the image color.
[0154] Noise Suppression:
[0155] The noise suppression sub-model is used to remove random noise and color noise in the image, and the clarity and quality of the image are improved through sophisticated algorithm processing.
[0156] Dynamic Range Adjustment:
[0157] Finally, the dynamic range adjustment sub-model is applied to optimize the bright and dark parts of the image so that the image details can be better displayed under different brightness conditions.
[0158] Beneficial Effects
[0159] Improve image quality: By comprehensively applying multiple sub-models to pre-process spectral images, the geometric accuracy, color authenticity, clarity and dynamic range of the images are significantly improved, providing high-quality image data for subsequent analysis.
[0160] Enhance the accuracy of subsequent analysis: Accurately corrected images can more truly reflect the actual situation of the monitored object, thereby improving the accuracy and reliability of subsequent analysis such as material composition analysis and state change monitoring.
[0161] Another embodiment of the image processing method based on AI recognition in the embodiment of the present invention includes: performing multi-level feature extraction on the spectral image after image preprocessing based on the preset sparse representation algorithm to obtain multi-level features of the target object, including:
[0162] Based on the preset spectral image segmentation rules, the preprocessed spectral image is segmented to obtain corresponding image blocks;
[0163] Based on a preset sparse representation algorithm, sparse encoding is performed on each image block to obtain a sparse representation result of each image block;
[0164] Based on the sparse representation result of each image block, hierarchical feature extraction is performed on the preprocessed spectral image to obtain multi-level features of the target object.
[0165] Specific, important term explanation:
[0166] Sparse representation algorithm: A signal processing method that can effectively capture the intrinsic structure of the data by finding the minimum number of primitives in an appropriate dictionary to represent the signal.
[0167] Spectral image segmentation rules: Preset rules are used to divide the spectral image into multiple regions or blocks for more detailed analysis and processing.
[0168] Multi-level feature extraction: A deep learning technique that reveals the deep information of data by extracting features of complex data layer by layer, from basic image features to more advanced abstract features.
[0169] Application scenarios:
[0170] This technical solution is suitable for scenarios that require in-depth analysis of spectral images, such as remote sensing image analysis, biomedical imaging, precision agriculture, environmental monitoring, etc. In these fields, being able to accurately extract multi-level features of target objects is crucial for recognition and classification tasks.
[0171] Furthermore, taking the analysis of vegetation coverage in environmental monitoring as an example, the researchers used this technical solution to process and analyze the collected vegetation spectral images:
[0172] Spectral image segmentation:
[0173] Firstly, according to the preset spectral image segmentation rules, the preprocessed vegetation spectral image is divided into a plurality of image blocks, each of which represents a specific area of vegetation.
[0174] Sparse Coding:
[0175] Then, a sparse representation algorithm is applied to each image block for sparse coding, and a sparse representation result of each image block is obtained by selecting a few primitives that are most suitable for representing the image block.
[0176] Multi-level feature extraction:
[0177] Based on the sparse representation results, the preprocessed spectral image is further subjected to hierarchical feature extraction to obtain multi-level features from shallow to deep layers, revealing deep information such as the health status and growth cycle of vegetation.
[0178] Beneficial effects:
[0179] Deep understanding of the target object: Through multi-level feature extraction, we can have a deeper understanding of the intrinsic attributes and status of the target object, such as the growth status and moisture content of vegetation.
[0180] Improve the accuracy of recognition and classification: The sparse features obtained by the sparse representation algorithm combined with multi-level analysis improve the accuracy and efficiency of target object recognition and classification tasks.
[0181] Optimize data processing flow: Integrate spectral image segmentation, sparse coding and multi-level feature extraction into one processing flow, which greatly improves the efficiency and effect of data processing.
[0182] Support for applications in complex environments: This technical solution can adapt to different application scenarios and complex environmental conditions, providing powerful image analysis tools for a wide range of fields.
[0183] Another embodiment of the image processing method based on AI recognition in the embodiment of the present invention includes: the training process of the adaptive detection model includes:
[0184] Collect training data and input the training data into a preliminarily constructed adaptive deep learning network; wherein the adaptive deep learning network includes at least a shape feature analysis sub-model, a texture analysis sub-model, a color recognition sub-model, a motion tracking sub-model, and a background separation sub-model; and the training data is a multi-level feature of the target object;
[0185] Based on the shape feature analysis sub-model, the contour features and shape features of the target object in the spectral image are analyzed to identify the basic structure of the target object and output the shape feature index;
[0186] Based on the texture analysis sub-model, the texture features of the surface of the target object in the spectral image are analyzed and a texture feature index is output; wherein the texture features at least include the roughness, directionality and repetitive pattern of the texture;
[0187] Based on the color recognition sub-model, the color characteristics and color distribution characteristics of the target object in the spectral image are identified and analyzed, the contribution of color to the recognition of the target object is evaluated, and the color characteristic index is output;
[0188] Based on the motion tracking sub-model, the dynamic image sequence characteristics of the target object in the spectral image are analyzed, the motion trajectory and speed of the target object are tracked, the motion pattern is analyzed, and the motion characteristic index is output;
[0189] Based on the background separation sub-model, analyze the surrounding environment background features of the target object in the spectral image, separate the target object from the surrounding environment background features, and output a background separation effect index;
[0190] Extracting a preset actual index value from the collected image data; wherein the actual index value represents the characteristic performance of the target object under an ideal state and is used as a reference standard in the training process;
[0191] Calculate an error value based on the deviation between the predicted index output by each sub-model in the constructed adaptive deep learning network and the preset actual index value; wherein the error value reflects the difference between the predicted index and the actual index value;
[0192] Based on the preset adaptive learning rate adjustment strategy, the parameters of each sub-model in the adaptive deep learning network are gradually adjusted to minimize the error value until the training of the adaptive deep learning network is completed to obtain the trained adaptive detection model.
[0193] Specific, important term explanation:
[0194] Adaptive deep learning network: A deep learning network model that can automatically adjust its structure and parameters according to the characteristics of the input data, aiming to improve the adaptability and performance of the model for different tasks and data.
[0195] Shape Feature Parsing Sub-Model: A deep learning sub-model that focuses on analyzing and understanding the shape and contour features of target objects in images.
[0196] Texture analysis sub-model: A sub-model used to parse the surface texture of target objects in the image and identify different texture patterns and characteristics.
[0197] Color recognition sub-model: A sub-model that recognizes and analyzes color features in an image, including color distribution, intensity, and saturation.
[0198] Motion tracking sub-model: A sub-model that tracks and analyzes the motion trajectory, speed, and motion pattern of the target object in the image sequence.
[0199] Background separation sub-model: A sub-model that separates the target object from the complex background, emphasizing the distinction between the target and the background.
[0200] Application scenarios:
[0201] This technical solution is widely applicable to security monitoring, intelligent transportation, medical image analysis, environmental monitoring and other fields, especially in situations where accurate identification and analysis of target objects are required, such as automatic identification of abnormal behavior in surveillance videos, vehicle identification in intelligent transportation systems, and detection of diseased tissue in medical images.
[0202] Furthermore, in the intelligent traffic monitoring system, the development team designed an adaptive detection model to accurately identify and track vehicles on the highway and separate vehicle images from complex backgrounds to achieve real-time traffic flow monitoring and accident warning.
[0203] Training data collection:
[0204] The team collected a large amount of vehicle spectral image data from traffic surveillance cameras in different cities, including images of different time periods, different weather conditions, and different traffic densities.
[0205] Adaptive deep learning network construction:
[0206] An adaptive deep learning network was designed which includes five key sub-models: shape feature analysis, texture analysis, color recognition, motion tracking, and background separation.
[0207] Feature analysis and indicator extraction:
[0208] The shape feature analysis sub-model is used to identify the basic outline and structure of the vehicle, the texture analysis sub-model analyzes the texture details of the vehicle surface, the color recognition sub-model determines the color characteristics of the vehicle, the motion tracking sub-model tracks the movement trajectory of the vehicle, and the background separation sub-model accurately separates the vehicle from the cluttered background.
[0209] Error calculation and model optimization:
[0210] The predicted indicators are compared with the actual indicator values extracted from the expert-annotated image data, and the error value is calculated. Then, based on the adaptive learning rate adjustment strategy, the network parameters are dynamically adjusted to minimize the error value and gradually optimize the model performance.
[0211] Beneficial effects:
[0212] Improved recognition accuracy: By comprehensively applying multi-dimensional analysis of shape, texture, color, motion trajectory and background separation, the model greatly improves the accuracy and robustness of vehicle recognition, especially under complex backgrounds and changing environmental conditions.
[0213] Real-time performance optimization: The adaptive deep learning network dynamically adjusts parameters to adapt to different monitoring scenarios, ensuring efficient data processing and fast recognition speed to meet the needs of real-time traffic monitoring.
[0214] Reduced false alarm rate: The effectiveness of the background separation sub-model significantly reduces the false alarm rate caused by background noise and enhances the system's ability to monitor real traffic conditions.
[0215] Strong system adaptability: The adaptive detection model can automatically adjust the learning strategy to adapt to different traffic environments and conditions, which improves the universal applicability and reliability of the system and is suitable for deployment in changing traffic monitoring scenarios.
[0216] Another embodiment of the image processing method based on AI recognition in the embodiment of the present invention includes: the actual index value includes at least actual contour features, actual shape features, actual texture features, actual color features, actual color distribution features, actual dynamic image sequence features, and actual surrounding environment background features.
[0217] The above describes the image processing method based on AI recognition in the embodiment of the present invention. The following describes the image processing system based on AI recognition in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an image processing system based on AI recognition includes:
[0218] An acquisition module is used to acquire images of dynamically changing scenes through an AI imaging device with a built-in multispectral sensor to obtain spectral images of dynamically changing scenes;
[0219] An image preprocessing module, used for inputting the spectral image into a preset physical model for image preprocessing to obtain a spectral image after image preprocessing;
[0220] An extraction module is used to perform multi-level feature extraction on the spectral image after image preprocessing based on a preset sparse representation algorithm to obtain multi-level features of the target object; wherein the target object includes at least crops, water bodies, and road traffic; the multi-level features of the target object include at least contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features, and surrounding environment background features of the target object;
[0221] A detection module, used for inputting the multi-level features of the target object into the trained adaptive detection model for real-time detection to obtain the multi-level feature detection results of the target object; wherein the adaptive detection model is obtained through pre-training;
[0222] The management module is used to analyze the multi-level feature detection results, obtain analysis results, and make decision management on the target object based on the analysis results.
[0223] The present invention also provides an image processing device based on AI recognition, which includes a memory and a processor, wherein computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the processor executes the steps of the image processing method based on AI recognition in the above-mentioned embodiments.
[0224] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the image processing method based on AI recognition.
[0225] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0226] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.
[0227] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An image processing method based on AI recognition, It is characterized in that The following steps are involved: The images of dynamically changing scenes are collected by an AI imaging device with a built-in multispectral sensor to obtain spectral images of dynamically changing scenes; Inputting the spectral image into a preset physical model for image preprocessing to obtain a spectral image after image preprocessing; Based on a preset sparse representation algorithm, multi-level feature extraction is performed on the spectral image after image preprocessing to obtain multi-level features of the target object; wherein the target object includes at least crops, water bodies, and road traffic; the multi-level features of the target object include at least contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features, and surrounding environment background features of the target object; The multi-level features of the target object are input into the trained adaptive detection model for real-time detection to obtain the multi-level feature detection results of the target object; wherein the adaptive detection model is obtained through advance training; the training process of the adaptive detection model includes: collecting training data, and inputting the training data into a preliminarily constructed adaptive deep learning network; wherein the adaptive deep learning network includes at least a shape feature analysis sub-model, a texture analysis sub-model, a color recognition sub-model, a motion tracking sub-model, and a background separation sub-model; the training data is the multi-level features of the target object; based on the shape feature analysis sub-model, the contour features and shape features of the target object in the spectral image are analyzed to identify the basic structure of the target object, and the shape feature index is output; based on the texture analysis sub-model, the texture features of the surface of the target object in the spectral image are analyzed, and the texture feature index is output; wherein the texture features include at least the roughness, directionality and repetitive pattern of the texture; based on the color recognition sub-model, the color features and color distribution features of the target object in the spectral image are identified and analyzed, Evaluate the contribution of color to the recognition of the target object and output the color feature index; based on the motion tracking sub-model, analyze the dynamic image sequence features of the target object in the spectral image, track the motion trajectory and speed of the target object, analyze the motion pattern, and output the motion feature index; based on the background separation sub-model, analyze the surrounding environment background features of the target object in the spectral image, separate the target object from the surrounding environment background features, and output the background separation effect index; extract the preset actual index value from the collected image data; wherein the actual index value represents the characteristic performance of the target object under an ideal state and is used as a reference standard in the training process; calculate the error value according to the deviation between the predicted index output by each sub-model in the constructed adaptive deep learning network and the preset actual index value; wherein the error value reflects the difference between the predicted index and the actual index value; based on the preset adaptive learning rate adjustment strategy, gradually adjust the parameters of each sub-model in the adaptive deep learning network to minimize the error value until the training of the adaptive deep learning network is completed to obtain the trained adaptive detection model; The multi-level feature detection results are analyzed to obtain analysis results, and decision management is performed on the target object based on the analysis results.
2. The image processing method based on AI recognition according to claim 1, It is characterized in that The method of collecting images of dynamically changing scenes by using an AI imaging device with a built-in multispectral sensor to obtain a spectral image of the dynamically changing scene includes: Automatically adjust the spectral response range of AI imaging devices with built-in multispectral sensors under different ambient lighting conditions to obtain optimized spectral image acquisition parameters; Based on the optimized spectral image acquisition parameters, the image of the dynamically changing scene is preliminarily acquired by an AI imaging device with a built-in multispectral sensor to obtain spectral features within a specific wavelength range; Applying a preset spectral analysis algorithm to extract spectral features within a specific wavelength range, identifying spectral features in each frame, and obtaining multi-frame spectral features; wherein the spectral features in each frame are used to represent material composition and state changes in a dynamically changing scene; The spectral features of multiple frames collected continuously are integrated to construct a spectral image of a dynamically changing scene; wherein the spectral image of the dynamically changing scene is used to represent the change trajectory of the spectral information of the target object in the dynamically changing scene over time.
3. The image processing method based on AI recognition according to claim 1, It is characterized in that The step of inputting the spectral image into a preset physical model for image preprocessing to obtain the spectral image after image preprocessing comprises: Obtain a preset physical model, and input the spectral image into the preset physical model for image preprocessing; wherein the preset physical model at least includes an optical distortion correction sub-model, a color deviation adjustment sub-model, a noise suppression sub-model, and a dynamic range adjustment sub-model; Analyzing the radial distortion and tangential distortion of the spectral image by using the optical distortion correction sub-model, calculating a first correction parameter corresponding to the radial distortion and a second correction parameter corresponding to the tangential distortion, and adjusting the position of each pixel in the spectral image according to the first correction parameter and the second correction parameter; Adjusting the color balance of the spectral image and compensating for the color deviation of the spectral image by using the color deviation adjustment sub-model; Removing random noise and color noise in the spectral image by using the noise suppression sub-model; The dynamic range of the spectral image is adjusted by the dynamic range adjustment sub-model to balance the bright and dark parts of the image; The spectral image is processed by the optical distortion correction sub-model, the color deviation adjustment sub-model, the noise suppression sub-model and the dynamic range adjustment sub-model respectively to obtain a spectral image after image preprocessing.
4. The image processing method based on AI recognition according to claim 1, It is characterized in that The method of performing multi-level feature extraction on the spectral image after image preprocessing based on the preset sparse representation algorithm to obtain multi-level features of the target object includes: Based on the preset spectral image segmentation rules, the preprocessed spectral image is segmented to obtain corresponding image blocks; Based on a preset sparse representation algorithm, sparse encoding is performed on each image block to obtain a sparse representation result of each image block; Based on the sparse representation result of each image block, hierarchical feature extraction is performed on the preprocessed spectral image to obtain multi-level features of the target object.
5. The image processing method based on AI recognition according to claim 1, It is characterized in that The actual index value includes at least actual contour features, actual shape features, actual texture features, actual color features, actual color distribution features, actual dynamic image sequence features, and actual surrounding environment background features.
6. An image processing system based on AI recognition, It is characterized in that The image processing system based on AI recognition includes: An acquisition module is used to acquire images of dynamically changing scenes through an AI imaging device with a built-in multispectral sensor to obtain spectral images of dynamically changing scenes; An image preprocessing module, used for inputting the spectral image into a preset physical model for image preprocessing to obtain a spectral image after image preprocessing; An extraction module is used to perform multi-level feature extraction on the spectral image after image preprocessing based on a preset sparse representation algorithm to obtain multi-level features of the target object; wherein the target object includes at least crops, water bodies, and road traffic; the multi-level features of the target object include at least contour features, shape features, texture features, color features, color distribution features, dynamic image sequence features, and surrounding environment background features of the target object; A detection module is used to input the multi-level features of the target object into a trained adaptive detection model for real-time detection to obtain a multi-level feature detection result of the target object; wherein the adaptive detection model is obtained through advance training; the training process of the adaptive detection model includes: collecting training data, and inputting the training data into a preliminarily constructed adaptive deep learning network; wherein the adaptive deep learning network at least includes a shape feature parsing sub-model, a texture analysis sub-model, a color recognition sub-model, a motion tracking sub-model, and a background separation sub-model; the training data is the multi-level features of the target object; based on the shape feature parsing sub-model, the contour features and shape features of the target object in the spectral image are analyzed to identify the basic structure of the target object, and the shape feature index is output; based on the texture analysis sub-model, the texture features of the surface of the target object in the spectral image are analyzed, and the texture feature index is output; wherein the texture features include at least the roughness, directionality and repetitive pattern of the texture; based on the color recognition sub-model, the color features and color separation of the target object in the spectral image are identified and analyzed The invention discloses a method for distributing features, evaluating the contribution of color to the recognition of the target object, and outputting a color feature index; based on the motion tracking sub-model, analyzing the dynamic image sequence features of the target object in the spectral image, tracking the motion trajectory and speed of the target object, analyzing the motion pattern, and outputting a motion feature index; based on the background separation sub-model, analyzing the surrounding environment background features of the target object in the spectral image, separating the target object from the surrounding environment background features, and outputting a background separation effect index; extracting a preset actual index value from the collected image data; wherein the actual index value represents the characteristic performance of the target object under an ideal state and is used as a reference standard in the training process; calculating an error value based on the deviation between the predicted index output by each sub-model in the constructed adaptive deep learning network and the preset actual index value; wherein the error value reflects the difference between the predicted index and the actual index value; based on the preset adaptive learning rate adjustment strategy, gradually adjusting the parameters of each sub-model in the adaptive deep learning network to minimize the error value until the training of the adaptive deep learning network is completed to obtain a trained adaptive detection model; The management module is used to analyze the multi-level feature detection results, obtain analysis results, and make decision management on the target object based on the analysis results.
7. An image processing device based on AI recognition, It is characterized in that The image processing device based on AI recognition includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the AI recognition-based image processing device executes the AI recognition-based image processing method according to any one of claims 1 to 5.
8. A computer-readable storage medium having instructions stored thereon, It is characterized in that When the instructions are executed by the processor, the image processing method based on AI recognition as described in any one of claims 1 to 5 is implemented.
