Image recognition system based on machine vision

Abstract

The invention discloses an image recognition system based on machine vision, which comprises a deep learning model, and is characterized in that an image recognition module comprises an image data sensing module for receiving and processing original image data, capturing an image from a camera or an image file, and performing preliminary format conversion and preprocessing; the image feature extraction module is used for converting the image into a form which can be understood by a machine based on a convolutional neural network; the online learning framework is used for dynamically adjusting model parameters according to data input in real time; the environment adaptability adjusting module is used for dynamically adjusting image processing parameters, changing a feature extraction weight, adjusting a learning rate and adjusting an identification strategy according to environment changes; the recognition and classification module is used for executing object recognition and classification tasks based on the extracted features and the learned model; the parameter mapping unit is used for mapping the environment parameters to specific parameter adjustment of the model; and the environment adaptability development unit is used for extracting useful information from the environment data and converting the useful information into an instruction of an adaptability adjustment strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision, and in particular to an image recognition system based on machine vision. Background Art

[0002] The massive amount of image data generated by the Internet, social media, and surveillance systems has provided rich training resources for image recognition technology. Data-driven has become a key factor in the success of deep learning. At the same time, with the widespread use of GPUs and the popularity of cloud computing services, it has become possible to train large-scale neural networks, reducing costs and accelerating the iteration and application of algorithms.

[0003] Now, in addition to the initial image classification and target detection to the current semantic segmentation, instance segmentation, posture estimation, action recognition, etc., image recognition technology is also widely used in security, medical, autonomous driving, retail, agriculture, and education. With the emergence of high-performance graphics processors and their application in general computing, it provides strong hardware support for large-scale parallel computing, making it possible to train complex deep learning models.

[0004] After searching, the invention patent with Chinese patent number CN117315377B discloses an image processing method, device and electronic device based on machine vision. Compared with the existing technology, the invention patent with Chinese patent number CN117315377B solves the problem of insufficient training data, and also performs noise reduction in the process of image recognition, thereby improving the recognition performance of the image recognition model and improving the accuracy of classification.

[0005] However, in the above-mentioned use process, while the image recognition process is being optimized, the machine learning image recognition effect will also be affected by external factors. The machine vision system has high requirements for the working environment. Light intensity, color, angle, shadow, temperature, and humidity will affect the image quality, thereby affecting the recognition accuracy. Under constantly changing natural light or inclement weather conditions, the recognition effect of the outdoor monitoring system may be greatly reduced. Therefore, an image recognition system based on machine vision is proposed. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art that the machine vision system has high requirements on the working environment and the image recognition effect is affected by external factors, and to propose an image recognition system based on machine vision.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] An image recognition system based on machine vision includes a deep learning model, and the image recognition module includes:

[0009] Image data perception module: receives and processes raw image data, captures images from cameras or image files, performs preliminary format conversion and preprocessing, and passes the image data after preliminary format conversion and preprocessing to the image feature extraction module;

[0010] Image feature extraction module: Based on the convolutional neural network (CNN), the image is converted into a machine-understandable form, and the key features in the image are extracted. The features are passed to the recognition and classification module, and the features are used to update the model of the online learning framework;

[0011] Online learning framework: dynamically adjusts model parameters based on real-time input data, and passes the updated model parameters to the image feature extraction module and the recognition and classification module;

[0012] Environmental adaptability adjustment module: This module adjusts the recognition strategy according to environmental changes by dynamically adjusting image processing parameters, changing feature extraction weights, and adjusting learning rates. The adjusted image processing parameters, feature extraction weights, and learning rates are passed to the image data perception module, image feature extraction module, and online learning framework, respectively.

[0013] Recognition and classification module: performs object recognition and classification tasks based on the extracted features and learned models;

[0014] Parameter mapping unit: maps environmental parameters to specific parameter adjustments of the model. The specific model parameter adjustment instructions are passed to the online learning framework and image feature extraction module;

[0015] Environmental adaptability development unit: extracts useful information from environmental data and converts it into instructions for adaptive adjustment strategies to achieve dynamic mapping and optimization of model parameters. The instructions for the adaptive adjustment strategies are passed to the parameter mapping unit and the environmental adaptability adjustment module.

[0016] Furthermore, the image data perception module monitors the light intensity of the current environment in real time based on the light intensity sensor;

[0017] Integrate temperature, humidity, wind speed, rainfall sensors, and image analysis algorithms to identify weather conditions;

[0018] Dynamically normalize the sensor data:

[0019]

[0020] rolling_min(X,w) and rolling_max(X,w) represent the minimum and maximum values ​​in the past w time windows. After processing, adaptive histogram equalization is performed considering the lighting conditions.

[0021] Furthermore, the image data perception module adjusts the brightness and contrast of the image based on histogram equalization, optimizes the illumination adaptability of histogram equalization, and Dynamically adjust image brightness and contrast;

[0022] Among them, n and N are the basic components of histogram equalization;

[0023] N is the total number of pixels, n is the number of pixels at a certain gray level;

[0024] p(i) represents the proportion of pixels with gray level i in the original image, that is, p(i), which is the basis for calculating the new gray value of each pixel in histogram equalization;

[0025] f(L) represents a function adjusted according to the current light intensity L;

[0026] When the light is strong, f(L) is designed to decrease to avoid excessive brightness enhancement;

[0027] When the light is weak, f(L) increases to enhance the visibility and contrast of the image. After the adjustment is completed, the image feature extraction module is activated.

[0028] Furthermore, the image feature extraction module collects the light intensity and weather information of the current environment through the environment perception module. At the front end of the convolutional neural network architecture, a special fusion layer is designed to combine the environmental features, normalized light intensity, weather type code and image pixel data. The environmental features are spliced ​​with the image features or combined through a specific function. The convolution kernel weights of the environment adaptation layer are designed to be dynamically adjusted according to the environmental features.

[0029] Introducing a learnable weight factor that adjusts the weights of the filters in the convolutional layer according to the environmental input, making the filters more likely to capture features that are useful under current lighting or weather conditions;

[0030] Design special filters in the convolutional layer to deal with illumination changes;

[0031] Train the deep learning model and activate the online learning framework.

[0032] Furthermore, the online learning framework introduces an environmentally aware reinforcement learning model, which maps the current environmental state, such as lighting and weather information, to a predefined recognition strategy library according to the environmental state. For example, a preprocessing step to enhance contrast may be used under low-light conditions, while image defogging may be added on rainy days. The image preprocessing or feature extraction strategy is adjusted in real time according to the environmental state to adapt to environmental changes;

[0033] Select an SVM model that supports incremental learning, calculate the accuracy, recall, and F1 score indicators in real time, and monitor the performance fluctuations of the model; design a dynamic adjustment strategy based on performance indicators to trigger model updates or parameter adjustments when the model performance is lower than the preset threshold or shows obvious degradation;

[0034] Evaluate the value of samples based on the current uncertainty or error rate of the model, actively select samples with the most uncertainty or the highest error rate for manual labeling, add the newly labeled samples to the training set, and update the model through incremental learning;

[0035] Establish a fast feedback mechanism to compare the model output with the actual label. Based on the feedback results, quickly adjust the model parameters or learning strategy, such as adjusting the learning rate and regularization strength, to form a closed-loop control.

[0036] Finally, the environmental adaptability adjustment module is activated, and the output of the environmental perception module is used as part of the model input or as the basis for adjusting parameters.

[0037] Furthermore, the environment adaptability adjustment module is based on the online learning framework, performs online training based on the newly received image data and labels, and updates the model parameters. The environment information provided by the environment adaptability adjustment module is used as an additional input or constraint to guide the parameter adjustment direction of the online learning;

[0038] The environmental adaptability adjustment module optimizes the adaptive learning rate method based on the online learning framework:

[0039]

[0040] m t and v t They are the first-order and second-order moment estimates, β1 and β2 are momentum terms to avoid zero division anomalies; the update rules of the two momentum terms correspond to the first-order moment mean momentum m, respectively. t and the second moment variance momentum v t ,in, Represents the model parameter w at time step t t About input x t and target y t The loss function gradient of , β1 and β2 are momentum decay rates, close to but less than 1, giving higher weight to recent gradients;

[0041] The environmental adaptability adjustment module is based on the accumulated momentum term and updates the model parameters w t :

[0042]

[0043] After learning the adaptive adjustment, the parameter mapping unit is activated.

[0044] Further, the parameter mapping unit defines a set of mapping rules or functions to map the environmental characteristic light intensity and weather type to the adjustment of the internal parameters of the model;

[0045] Design a mapping function between light intensity and image preprocessing gain parameters, design adjustable activation function thresholds for specific layers in the model, define a weather-type-based mapping table or function to map different weather types (sunny, cloudy, rainy) to different threshold settings. For example, a rainy day may require a lower threshold to capture more subtle feature changes. Build a deep learning model with adjustable parameter layers, learnable weights or bias terms, and dynamically adjust parameters based on environmental inputs. Define a set of mapping rules or functions to map environmental feature light intensity and weather type to the adjustment of model internal parameters, and activate the environmental adaptability development unit;

[0046] Dynamic threshold adjustment of the activation function by the environmental adaptability development unit;

[0047] Collect a large amount of historical data, including images and corresponding labels under different environmental conditions, analyze this data, determine which activation function threshold settings can bring the best model performance under different environmental conditions, and use statistical methods to establish a mapping model between thresholds and environmental characteristics. When new environmental data is input, first obtain the current light intensity and weather type through the environmental perception unit, then calculate the corresponding activation function threshold according to the mapping model and apply it to the model;

[0048] Online learning technology is used to adjust the threshold of the activation function according to the newly received environmental features and model performance feedback, and the environment is divided into several categories: sunny, cloudy, and rainy. Different activation threshold strategies are learned or set for each category separately, and dynamically switched according to the real-time environmental classification.

[0049] Furthermore, the recognition and classification module performs the image classification task based on the adjusted model parameters;

[0050] Generate classification results, including the identified object category, location, and confidence information.

[0051] The present invention has the following beneficial effects:

[0052] 1. In the present invention, by adjusting the model parameters in real time, the system can effectively cope with changes in light intensity and adverse weather conditions, reduce the negative impact of external environmental factors on recognition accuracy, and improve the robustness and stability of image recognition. Through the environmental adaptability adjustment module, the system can dynamically adjust image processing parameters, feature extraction weights and learning rates to cope with changes in external factors such as light intensity, color, angle, shadow, etc., which helps to reduce the degradation of image quality caused by environmental factors, thereby improving recognition accuracy. The environmental adaptability development unit can extract useful information from environmental data and convert it into instructions for adaptive adjustment strategies, which enables the system to actively adapt to changes in the external environment rather than just passively respond. This active adaptability helps the system maintain stable recognition performance when facing complex and changing environments.

[0053] 2. In the present invention, the performance of the model in a specific environment can be optimized by dynamically adjusting the hyperparameters and activation function thresholds, so that the model can more accurately capture and distinguish key features in the image, especially in complex or low-quality image scenes, significantly improving the recognition accuracy. At the same time, the environmental parameters are mapped to the specific parameter adjustment of the model, realizing the dynamic mapping and optimization of the model parameters. This helps the system to more accurately adjust its internal parameters to match the current working environment, thereby further improving the recognition effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a system block diagram of an image recognition system based on machine vision proposed by the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Embodiment 1

[0057] like Figure 1 As shown, the present invention proposes an image recognition system based on machine vision, including a deep learning model, and the image recognition module includes:

[0058] Image data perception module: receives and processes raw image data, captures images from cameras or image files, performs preliminary format conversion and preprocessing, and passes the image data after preliminary format conversion and preprocessing to the image feature extraction module;

[0059] Image feature extraction module: Based on the convolutional neural network (CNN), the image is converted into a machine-understandable form, and the key features in the image are extracted. The features are passed to the recognition and classification module, and the features are used to update the model of the online learning framework;

[0060] Online learning framework: dynamically adjusts model parameters based on real-time input data, and passes the updated model parameters to the image feature extraction module and the recognition and classification module;

[0061] Environmental adaptability adjustment module: This module adjusts the recognition strategy according to environmental changes by dynamically adjusting image processing parameters, changing feature extraction weights, and adjusting learning rates. The adjusted image processing parameters, feature extraction weights, and learning rates are passed to the image data perception module, image feature extraction module, and online learning framework, respectively.

[0062] Recognition and classification module: performs object recognition and classification tasks based on the extracted features and learned models;

[0063] Parameter mapping unit: maps environmental parameters to specific parameter adjustments of the model. The specific model parameter adjustment instructions are passed to the online learning framework and image feature extraction module;

[0064] Environmental adaptability development unit: extracts useful information from environmental data and converts it into instructions for adaptive adjustment strategies to achieve dynamic mapping and optimization of model parameters. The instructions for the adaptive adjustment strategies are passed to the parameter mapping unit and the environmental adaptability adjustment module.

[0065] The image data perception module is based on a light intensity sensor to monitor the light intensity of the current environment in real time;

[0066] Integrate temperature, humidity, wind speed, rainfall sensors, and image analysis algorithms to identify weather conditions;

[0067] Dynamically normalize the sensor data:

[0068]

[0069] rolling_min(X,w) and rolling_max(X,w) represent the minimum and maximum values ​​in the past w time windows. After processing, adaptive histogram equalization is performed considering the lighting conditions.

[0070] The image data perception module adjusts the brightness and contrast of the image based on histogram equalization and optimizes the illumination adaptability of histogram equalization. Dynamically adjust image brightness and contrast;

[0071] Among them, n and N are the basic components of histogram equalization;

[0072] N is the total number of pixels, n is the number of pixels at a certain gray level;

[0073] p(i) represents the proportion of pixels with gray level i in the original image, that is, p(i), which is the basis for calculating the new gray value of each pixel in histogram equalization;

[0074] f(L) represents a function adjusted according to the current light intensity L;

[0075] When the light is strong, f(L) is designed to decrease to avoid excessive brightness enhancement;

[0076] When the light is weak, f(L) increases to enhance the visibility and contrast of the image. After the adjustment is completed, the image feature extraction module is activated.

[0077] The image feature extraction module collects the light intensity and weather information of the current environment through the environmental perception module. At the front end of the convolutional neural network architecture, a special fusion layer is designed to combine environmental features, normalized light intensity, weather type encoding and image pixel data. The environmental features are spliced ​​with image features or combined through specific functions. The convolution kernel weights of the environmental adaptation layer are designed to be dynamically adjusted according to the environmental features.

[0078] Introducing a learnable weight factor that adjusts the weights of the filters in the convolutional layer according to the environmental input, making the filters more likely to capture features that are useful under current lighting or weather conditions;

[0079] Design special filters in the convolutional layer to deal with illumination changes;

[0080] Train the deep learning model and activate the online learning framework.

[0081] The online learning framework introduces an environmental perception reinforcement learning model. Based on the current environmental state, such as lighting and weather information, it is mapped to a predefined recognition strategy library according to the environmental state. For example, a preprocessing step to enhance contrast may be used under low-light conditions, while image defogging may be added on rainy days. The image preprocessing or feature extraction strategy is adjusted in real time according to the environmental state to adapt to environmental changes.

[0082] Select an SVM model that supports incremental learning, calculate the accuracy, recall, and F1 score indicators in real time, and monitor the performance fluctuations of the model; design a dynamic adjustment strategy based on performance indicators to trigger model updates or parameter adjustments when the model performance is lower than the preset threshold or shows obvious degradation;

[0083] Evaluate the value of samples based on the current uncertainty or error rate of the model, actively select samples with the most uncertainty or the highest error rate for manual labeling, add the newly labeled samples to the training set, and update the model through incremental learning;

[0084] Establish a fast feedback mechanism to compare the model output with the actual label. Based on the feedback results, quickly adjust the model parameters or learning strategy, such as adjusting the learning rate and regularization strength, to form a closed-loop control.

[0085] Finally, the environmental adaptability adjustment module is activated, and the output of the environmental perception module is used as part of the model input or as the basis for adjusting parameters.

[0086] In this embodiment,

[0087] The rolling_min(X,w) and rolling_max(X,w) represent the maximum and minimum values ​​of the past w data points (within the time window). The purpose is to allow the normalization process to take into account the dynamic changes of the data, rather than just the global maximum and minimum values ​​of the entire data set. Therefore, it is more suitable for dealing with environmental variables that change over time, such as the fluctuation of light intensity over time.

[0088] The numerator in the formula is the data range in the current window (maximum value minus minimum value), and the denominator is the difference between the current value and the minimum value of the window. The relative position of the current value relative to the current window is calculated. In this way, the data is mapped to the interval [0,1], but at the same time, the change trend of the recent data is taken into account, making the data preprocessing more adapted to the current environmental status.

[0089] Embodiment 2

[0090] like Figure 1 As shown, based on the first embodiment, the environment adaptability adjustment module is based on the online learning framework, performs online training based on the newly received image data and labels, updates the model parameters, and the environment information provided by the environment adaptability adjustment module is used as an additional input or constraint condition to guide the parameter adjustment direction of the online learning;

[0091] The environmental adaptability adjustment module optimizes the adaptive learning rate method based on the online learning framework:

[0092]

[0093] m t and v t They are the first-order and second-order moment estimates, β1 and β2 are momentum terms to avoid zero division anomalies; the update rules of the two momentum terms correspond to the first-order moment mean momentum m, respectively. tand the second moment variance momentum v t ,in, Represents the model parameter w at time step t t About input x t and target y t The loss function gradient of , β1 and β2 are momentum decay rates, close to but less than 1, giving higher weight to recent gradients;

[0094] The environmental adaptability adjustment module is based on the accumulated momentum term and updates the model parameters w t :

[0095]

[0096] After learning the adaptive adjustment, the parameter mapping unit is activated.

[0097] The parameter mapping unit defines a set of mapping rules or functions to map the environmental characteristics of light intensity and weather type to the adjustment of the model's internal parameters;

[0098] Design a mapping function between light intensity and image preprocessing gain parameters, design adjustable activation function thresholds for specific layers in the model, define a weather-type-based mapping table or function to map different weather types (sunny, cloudy, rainy) to different threshold settings. For example, a rainy day may require a lower threshold to capture more subtle feature changes. Build a deep learning model with adjustable parameter layers, learnable weights or bias terms, and dynamically adjust parameters based on environmental inputs. Define a set of mapping rules or functions to map environmental feature light intensity and weather type to the adjustment of model internal parameters, and activate the environmental adaptability development unit;

[0099] Dynamic threshold adjustment of the activation function by the environmental adaptability development unit;

[0100] Collect a large amount of historical data, including images and corresponding labels under different environmental conditions, analyze this data, determine which activation function threshold settings can bring the best model performance under different environmental conditions, and use statistical methods to establish a mapping model between thresholds and environmental characteristics. When new environmental data is input, first obtain the current light intensity and weather type through the environmental perception unit, then calculate the corresponding activation function threshold according to the mapping model and apply it to the model;

[0101] Online learning technology is used to adjust the threshold of the activation function according to the newly received environmental features and model performance feedback, and the environment is divided into several categories: sunny, cloudy, and rainy. Different activation threshold strategies are learned or set for each category separately, and dynamically switched according to the real-time environmental classification.

[0102] The recognition and classification module performs image classification tasks based on the adjusted model parameters;

[0103] Generate classification results, including the identified object category, location, and confidence information.

[0104] In this embodiment, the online learning framework is responsible for updating the model parameters in real time according to the newly received data, optimizing the generalization ability of the model so that it can gradually adapt to changes in data distribution. The environmental adaptability adjustment module is specifically optimized for environmental changes such as light and weather, and improves the recognition effect in a specific environment by adjusting the model parameters or algorithm strategies;

[0105] First, the environmental adaptability adjustment module starts working, monitoring environmental parameters (such as light intensity and weather type) through sensors and performing necessary data preprocessing, such as normalization;

[0106] Next, the image is preprocessed by histogram equalization and noise filtering, and then the image features are extracted using a deep learning model. At this point, the environmental adaptability adjustment module may have affected the image preprocessing parameters or the initial settings of the deep learning model.

[0107] The online learning framework starts to operate, and the model is trained online based on the newly received image data and labels, and the model parameters are updated. The environmental information provided by the environmental adaptability adjustment module serves as an additional input or constraint to guide the parameter adjustment direction of online learning;

[0108] The performance of the model on new data is monitored in real time. The environmental adaptability adjustment module triggers further parameter fine-tuning or strategy adjustment based on the performance feedback of the model in the current environment, forming a closed-loop feedback mechanism.

[0109] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition system based on machine vision, including a deep learning model, characterized in that: The image recognition module includes: Image data perception module: receives and processes raw image data, captures images from cameras or image files, performs preliminary format conversion and preprocessing, and passes the image data after preliminary format conversion and preprocessing to the image feature extraction module; Image feature extraction module: Based on the convolutional neural network (CNN), the image is converted into a machine-understandable form, and the key features in the image are extracted. The features are passed to the recognition and classification module, and the features are used to update the model of the online learning framework; Online learning framework: dynamically adjusts model parameters based on real-time input data, and passes the updated model parameters to the image feature extraction module and the recognition and classification module; Environmental adaptability adjustment module: adjusts the recognition strategy according to environmental changes by dynamically adjusting image processing parameters, changing feature extraction weights, and adjusting learning rates. The adjusted image processing parameters, feature extraction weights, and learning rates are passed to the image data perception module, image feature extraction module, and online learning framework, respectively. Recognition and classification module: performs object recognition and classification tasks based on the extracted features and learned models; Parameter mapping unit: maps environmental parameters to specific parameter adjustments of the model. The specific model parameter adjustment instructions are passed to the online learning framework and image feature extraction module; Environmental adaptability development unit: extracts useful information from environmental data and converts it into instructions for adaptive adjustment strategies to achieve dynamic mapping and optimization of model parameters. The instructions for the adaptive adjustment strategies are passed to the parameter mapping unit and the environmental adaptability adjustment module.

2. The image recognition system based on machine vision according to claim 1, characterized in that: The image data perception module is based on a light intensity sensor to monitor the light intensity of the current environment in real time; Integrate temperature, humidity, wind speed, rainfall sensors, and image analysis algorithms to identify weather conditions; Dynamically normalize the sensor data: rolling_min(X,w) and rolling_max(X,w) represent the minimum and maximum values ​​in the past w time windows. After processing, adaptive histogram equalization is performed considering the lighting conditions.

3. The image recognition system based on machine vision according to claim 1, characterized in that: The image data perception module adjusts the brightness and contrast of the image based on histogram equalization and optimizes the illumination adaptability of histogram equalization. Dynamically adjust image brightness and contrast; Among them, n and N are the basic components of histogram equalization; N is the total number of pixels, n is the number of pixels at a certain gray level; p(i) represents the proportion of pixels with gray level i in the original image, that is, p(i), which is the basis for calculating the new gray value of each pixel in histogram equalization; f(L) represents a function adjusted according to the current light intensity L; When the light is strong, f(L) is designed to decrease to avoid excessive brightness enhancement; When the light is weak, f(L) increases to enhance the visibility and contrast of the image. After the adjustment is completed, the image feature extraction module is activated.

4. The image recognition system based on machine vision according to claim 1, characterized in that: The image feature extraction module collects the light intensity and weather information of the current environment through the environmental perception module. At the front end of the convolutional neural network architecture, a special fusion layer is designed to combine environmental features, normalized light intensity, weather type encoding and image pixel data. The environmental features are spliced ​​with image features or combined through specific functions. The convolution kernel weights of the environmental adaptation layer are designed to be dynamically adjusted according to the environmental features. Introducing a learnable weight factor that adjusts the weights of the filters in the convolutional layer according to the environmental input, making the filters more likely to capture features that are useful under current lighting or weather conditions; Design special filters in the convolutional layer to deal with illumination changes; Train the deep learning model and activate the online learning framework.

5. The image recognition system based on machine vision according to claim 1, characterized in that: The online learning framework introduces an environment-aware reinforcement learning model to adjust the recognition strategy based on the current environment state. Select an SVM model that supports incremental learning, calculate the accuracy, recall, and F1 score indicators in real time, and monitor the performance fluctuations of the model; design a dynamic adjustment strategy based on performance indicators to trigger model updates or parameter adjustments when the model performance is lower than the preset threshold or shows obvious degradation; Actively select the most valuable samples for labeling and learning based on the model’s current uncertainty or error rate; Establish a fast feedback mechanism to compare the model output with the actual label, and quickly adjust the model parameters or learning strategy through feedback to form a closed-loop control; Finally, activate the environmental adaptability adjustment module.

6. The image recognition system based on machine vision according to claim 1, characterized in that: The environment adaptability adjustment module is based on the online learning framework and performs online training based on the newly received image data and labels to update the model parameters. The environment information provided by the environment adaptability adjustment module is used as an additional input or constraint to guide the parameter adjustment direction of online learning. The environmental adaptability adjustment module optimizes the adaptive learning rate method based on the online learning framework: m t =β1m t-1 +(1-β1)▽L(w t ,x t ,y t ) v t =β2v t -1+(1-β2)(▽L(w t ,x t ,y t )) 2 Among them, m t and v t are the first-order and second-order moment estimates, respectively, β1 and β2 are momentum terms to avoid division by zero anomalies; The two momentum term update rules correspond to the first-order moment mean momentum m t and the second moment variance momentum v t ,in, Represents the model parameter w at time step t t About input x t and target y t The loss function gradient of , β1 and β2 are momentum decay rates, close to but less than 1, giving higher weight to recent gradients; The environmental adaptability adjustment module is based on the accumulated momentum term and updates the model parameters w t : After learning the adaptive adjustment, the parameter mapping unit is activated.

7. The image recognition system based on machine vision according to claim 1, characterized in that: The parameter mapping unit defines a set of mapping rules or functions to map the environmental characteristics of light intensity and weather type to the adjustment of the model's internal parameters; Design a mapping function between light intensity and image preprocessing gain parameters, design adjustable activation function thresholds for specific layers in the model, build a deep learning model with adjustable parameter layers, learnable weights or bias terms, and dynamically adjust parameters based on environmental inputs, define a set of mapping rules or functions to map environmental characteristic light intensity and weather type to the adjustment of model internal parameters, and activate environmental adaptability development units; Dynamic threshold adjustment of the activation function by the environmental adaptability development unit; Based on historical data, the optimal threshold of the activation function under different environmental conditions is analyzed, and a mapping model between the threshold and environmental characteristics is established; Online learning technology is used to adjust the threshold of the activation function according to the newly received environmental features and model performance feedback, and the environment is divided into several categories: sunny, cloudy, and rainy. Different activation threshold strategies are learned or set for each category separately, and dynamically switched according to the real-time environmental classification.

8. The image recognition system based on machine vision according to claim 1, characterized in that: The recognition and classification module performs image classification tasks based on the adjusted model parameters and generates classification results, including the identified object category, location, and confidence information.

Images