A visible light visual weather recognition method and system

Abstract

The present application belongs to the technical field of weather identification. A visible light vision weather identification method and system are disclosed. The method comprises the following steps: constructing a weather identification model according to a plurality of historical panoramic image data and corresponding historical geographical environment data of different geographical environment regions; acquiring real-time panoramic video data and corresponding real-time geographical environment data of a target region; frame intercepting the real-time panoramic video data to obtain real-time panoramic image data of continuous frames; using the weather identification model to perform weather identification according to the real-time panoramic image data of continuous frames and the corresponding real-time geographical environment data, and obtaining a weather identification result of the target region. The system comprises a visible light camera device and a weather identification center. The present application solves the problems of low accuracy, poor efficiency, poor timeliness and high cost investment in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of weather recognition technology, specifically relating to a visible light visual weather recognition method and system. Background Technology

[0002] Weather is a major concern for people, playing a crucial role in guiding their travel. Knowing weather changes in advance allows people to make timely arrangements and avoid disruptions to their lives and work. Severe weather conditions frequently lead to catastrophic events such as ship collisions, forest fires, power plant and grid outages, train derailments, and plane crashes.

[0003] Weather identification and forecasting can solve many problems, and a lot of experience in weather identification and forecasting has been passed down since ancient times. However, with the continuous improvement of people's living standards and the continuous development of science and technology, relying solely on experience to identify and forecast the weather often cannot meet people's needs. People need more accurate and intelligent methods for weather identification and forecasting.

[0004] Existing weather stations often rely on sensors (such as microwave sensors or satellite sensors) for identification, but this method mainly depends on the experience and knowledge of experts for analysis and judgment, resulting in low accuracy, poor efficiency, and poor timeliness. Furthermore, building weather stations in specific areas in the field requires a large amount of manpower and material resources, resulting in high costs. Summary of the Invention

[0005] In order to address the problems of low accuracy, poor efficiency, poor timeliness, and high cost of existing technologies, the present invention aims to provide a visible light visual weather recognition method and system.

[0006] The technical solution adopted in this invention is as follows:

[0007] A visible light visual weather recognition method includes the following steps:

[0008] A weather recognition model is constructed based on several historical panoramic image data and corresponding historical geographical environment data of different geographical regions.

[0009] Acquire real-time panoramic video data and corresponding real-time geographic environment data of the target area;

[0010] Frames are extracted from real-time panoramic video data to obtain continuous frames of real-time panoramic image data.

[0011] Based on real-time panoramic image data of consecutive frames and corresponding real-time geographic environment data, a weather recognition model is used to perform weather recognition and obtain the weather recognition results for the target area.

[0012] Furthermore, the data types of historical / real-time geographic environment data include time, latitude and longitude, wind speed, wind direction, temperature, humidity, altitude, light intensity, and fog value.

[0013] Furthermore, the weather recognition model includes an image semantic segmentation sub-model, a geographic environment recognition sub-model, and a weather recognition sub-model. The image semantic segmentation sub-model is connected to the geographic environment recognition sub-model and the weather recognition sub-model, respectively, and the geographic environment recognition sub-model is connected to the weather recognition sub-model.

[0014] Furthermore, the image semantic segmentation sub-model is constructed based on the GSCNN algorithm, and the image semantic segmentation sub-model includes a first input layer, a rule flow module, a shape flow module, a fusion module, and a first output layer. The first input layer is connected to the rule flow module and the shape flow module, respectively. The rule flow module and the shape flow module are both connected to the fusion module. The fusion module is connected to the first output layer, and the first output layer is connected to the geographic environment recognition sub-model and the weather recognition sub-model, respectively.

[0015] The geographic environment recognition sub-model is built based on the LFMCNN algorithm and includes a second input layer, a first convolutional module, a downsampling module, a second convolutional module, a third convolutional module, a fully connected layer, and a second output layer. The second input layer is connected to the first output layer, the first convolutional module, and the downsampling module of the image semantic segmentation sub-model. The first convolutional module and the downsampling module are both connected to the second convolutional module. The second convolutional module, the third convolutional module, the fully connected layer, and the second output layer are connected in sequence. The second output layer is connected to the weather recognition sub-model.

[0016] The weather recognition sub-model is built based on the ShuffleNet algorithm and includes a third input layer, a first group convolutional layer, a feature fusion layer, a Shuffle layer, a second group convolutional layer, and a third output layer. The third input layer is connected to the first output layer and the first group convolutional layer of the image semantic segmentation sub-model, respectively. The first group convolutional layer, the feature fusion layer, the Shuffle layer, the second group convolutional layer, and the third output layer are connected in sequence. The feature fusion layer is connected to the second output layer of the geographic environment recognition sub-model.

[0017] Furthermore, based on several historical panoramic image data and corresponding historical geographical environment data of different geographical regions, a weather recognition model is constructed, including the following steps:

[0018] Several historical panoramic image data from different geographical regions were preprocessed to obtain a preprocessed historical panoramic image dataset.

[0019] The preprocessed historical panoramic image set is divided into a historical image semantic segmentation training dataset, a historical geographic environment recognition training dataset, and a historical weather recognition training dataset.

[0020] Based on the historical image semantic segmentation training dataset, the GSCNN algorithm is used for optimization training to construct an image semantic segmentation sub-model and obtain the corresponding historical semantic segmentation image dataset.

[0021] By setting the grayscale values ​​of the non-interest regions in the historical semantic segmentation image dataset to 0, the corresponding historical geographic environment segmentation image dataset and historical sky segmentation image dataset are obtained.

[0022] The historical geographic environment segmentation image dataset is added to the historical geographic environment recognition training dataset to obtain the updated historical geographic environment recognition training dataset. The historical sky segmentation image dataset is added to the historical weather recognition training dataset to obtain the updated historical weather recognition training dataset.

[0023] Based on the updated historical geographic environment recognition training dataset, the LFMCNN algorithm is used for optimization training to construct a geographic environment recognition sub-model and obtain several corresponding geographic environment data weight arrays.

[0024] Based on the updated historical weather recognition training dataset, as well as several historical geographical environment data and several geographical environment data weight arrays, the ShuffleNet algorithm is used for optimization training to construct a weather recognition sub-model.

[0025] The image semantic segmentation sub-model, the geographic environment recognition sub-model, and the weather recognition sub-model are combined to construct a weather recognition model.

[0026] Furthermore, based on the real-time panoramic image data of consecutive frames and the corresponding real-time geographic environment data, a weather recognition model is used to perform weather recognition to obtain the weather recognition results for the target area, including the following steps:

[0027] Preprocessing and frame skipping filtering are performed on continuous real-time panoramic image data to obtain a preprocessed real-time panoramic image dataset containing several key frames.

[0028] The preprocessed real-time panoramic image data of the current keyframe is input into the image semantic segmentation sub-model to perform image semantic segmentation and obtain the corresponding real-time semantic segmentation image data.

[0029] By setting the grayscale value of the non-interest region of the real-time semantic segmentation image data to 0, the corresponding real-time geographic environment segmentation image data and real-time sky segmentation image data are obtained.

[0030] Real-time geographic environment segmentation image data is input into the geographic environment recognition sub-model to perform geographic environment recognition and obtain the corresponding geographic environment category and geographic environment data weight array.

[0031] The real-time sky segmentation image data of the current keyframe, along with the corresponding real-time geographic environment data and geographic environment data weight array, are input into the weather recognition sub-model to perform weather recognition and obtain the weather recognition result of the current keyframe.

[0032] By traversing all key frames of the preprocessed real-time panoramic image data in the preprocessed real-time panoramic image dataset, weather recognition results for several key frames are obtained.

[0033] The weather recognition results of the target area are output as the weather recognition results of the keyframes that belong to the same category and are the most numerous among several keyframes.

[0034] Furthermore, the preprocessing includes sequential image size unification, image enhancement, image grayscale processing, and grayscale value normalization.

[0035] A visible light visual weather recognition system is provided for implementing a visible light visual weather recognition method. The system includes a visible light camera device and a weather recognition center. The visible light camera device is communicatively connected to the weather recognition center. The visible light camera device is equipped with a visible light camera and a geographic environment data acquisition unit. The weather recognition center is equipped with a weather recognition model and is communicatively connected to an external data server.

[0036] The visible light camera device is used to collect real-time panoramic video data of a target area using a visible light camera, collect real-time geographic environment data of the target area using a geographic environment data acquisition unit, and upload the real-time panoramic video data and corresponding real-time geographic environment data of the target area to a weather recognition center.

[0037] The weather recognition center is used to acquire historical panoramic image data and corresponding historical geographical environment data of different geographical areas sent by external data servers; construct a weather recognition model based on the historical panoramic image data and corresponding historical geographical environment data of different geographical areas; extract frames from real-time panoramic video data to obtain continuous frames of real-time panoramic image data; and perform weather recognition using the weather recognition model based on the continuous frames of real-time panoramic image data and corresponding real-time geographical environment data to obtain the weather recognition result of the target area.

[0038] Furthermore, the visible light camera device includes a main control unit, a visible light camera, a geographic environment data acquisition unit, and a communication unit. The main control unit is electrically connected to the visible light camera, the geographic environment data acquisition unit, and the communication unit, respectively. The communication unit is communicatively connected to the weather identification center.

[0039] Furthermore, the weather recognition center includes a data acquisition unit, a frame capture unit, a model building unit, and a weather recognition unit. The data acquisition unit is connected to a visible light camera device and an external data server, and is also connected to the frame capture unit, the weather recognition unit, and the model building unit. Both the frame capture unit and the model building unit are connected to the weather recognition unit.

[0040] The beneficial effects of this invention are as follows:

[0041] This invention provides a visible light visual weather recognition method and system. By using panoramic video data acquired through visible light visual technology and image recognition technology, a weather recognition model is constructed for automated, real-time weather recognition. This avoids reliance on manual analysis and judgment, improving the accuracy and efficiency of weather recognition and ensuring its timeliness. Furthermore, by using geographic environmental data acquired through sensor technology, the impact of geographic environmental factors on weather recognition is considered, improving the adaptability and accuracy of the weather recognition method and system to special outdoor environments. This also avoids the need to build weather stations in specific areas of the field, reducing the cost of weather recognition.

[0042] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0043] Figure 1 This is a flowchart of the visible light visual weather recognition method in this invention.

[0044] Figure 2 This is a structural block diagram of the visible light visual weather recognition system in this invention. Detailed Implementation

[0045] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0046] Example 1:

[0047] like Figure 1 As shown, this embodiment provides a visible light visual weather recognition method, including the following steps:

[0048] A weather recognition model is constructed based on several historical panoramic image data and corresponding historical geographical environment data for different geographical environments. These different geographical environments include desert areas, rainforest areas, ridge areas, hillside areas, plain areas, grassland areas, rural areas, and urban areas. The historical geographical environment data includes historical time, historical latitude and longitude, historical wind speed, historical wind direction, historical temperature, historical humidity, historical altitude, historical light intensity, and historical fog value, which were collected simultaneously with the historical panoramic image data.

[0049] The weather recognition model includes an image semantic segmentation sub-model, a geographic environment recognition sub-model, and a weather recognition sub-model. The image semantic segmentation sub-model is connected to the geographic environment recognition sub-model and the weather recognition sub-model, respectively. The geographic environment recognition sub-model is also connected to the weather recognition model.

[0050] The image semantic segmentation sub-model is constructed based on the Gated-Shape Convolutional Neural Networks (GSCNN) algorithm. The image semantic segmentation sub-model includes a first input layer, a rule flow module, a shape flow module, a fusion module, and a first output layer. The first input layer is connected to the rule flow module and the shape flow module, respectively. The rule flow module and the shape flow module are both connected to the fusion module. The fusion module is connected to the first output layer. The first output layer is connected to the geographic environment recognition sub-model and the weather recognition sub-model, respectively.

[0051] The image semantic segmentation sub-model uses a classic convolutional neural network structure to process semantic region information in its rule flow module and a shape flow module composed of multiple gated convolutional layers. The shape flow module uses low-level feature maps in the rule flow module to process region boundary information. The outputs of the two modules are fed into a fusion module. In the fusion module, the outputs of the rule flow module and the shape flow module are combined using an Atrous Special Pyramid Pooling (ASPP) structure. ASPP helps to preserve multi-scale contextual information. Finally, the fusion module generates object semantic regions with defined boundaries, resulting in accurately segmented semantic image data.

[0052] The geographic environment recognition sub-model is constructed based on the Low-level Feature Fusion Multi-kernel Convolutional Neural Network (LFMCNN) algorithm. The geographic environment recognition sub-model includes a second input layer, a first convolutional module, a downsampling module, a second convolutional module, a third convolutional module, a fully connected layer, and a second output layer. The second input layer is connected to the first output layer, the first convolutional module, and the downsampling module of the image semantic segmentation sub-model. The first convolutional module and the downsampling module are both connected to the second convolutional module. The second convolutional module, the third convolutional module, the fully connected layer, and the second output layer are connected in sequence. The second output layer is connected to the weather recognition sub-model.

[0053] The geographic environment recognition sub-model consists of three convolutional modules and a fully connected layer. This model can obtain information from different scales of the input image, and then perform downsampling on the original image. It is then fused with the feature map obtained after the convolution operation of the first convolutional module to obtain more spatial, color, and contour information. After feature extraction by two more convolutional modules, the model is finally classified, which improves the accuracy of geographic environment recognition.

[0054] The weather recognition sub-model is built based on the ShuffleNet algorithm and includes a third input layer, a first group convolutional layer, a feature fusion layer, a Shuffle layer, a second group convolutional layer, and a third output layer. The third input layer is connected to the first output layer and the first group convolutional layer of the image semantic segmentation sub-model, respectively. The first group convolutional layer, the feature fusion layer, the Shuffle layer, the second group convolutional layer, and the third output layer are connected in sequence. The feature fusion layer is connected to the second output layer of the geographic environment recognition sub-model.

[0055] The weather recognition sub-model employs multiple grouped convolutional layers, which can reduce the computational load and parameter count by several times when the images have the same input and output dimensions, thereby achieving the goal of a lightweight network. The feature fusion layer multiplies the weight array of the geographic environment data with the input historical / real-time geographic environment data to differentiate the use of geographic environment data, and then extracts geographic environment data features. These geographic environment data features are then fused with the historical / real-time sky segmentation image data features extracted by the first grouped convolutional layer to obtain fused features. The Shuffle layer randomly shuffles the feature maps of the fused features, allowing information to flow between the groups in the second grouped convolutional layer without incurring any additional computational cost, thus improving the recognition efficiency and accuracy of the weather recognition sub-model. The third output layer outputs the corresponding weather recognition result. By taking into account the impact of the geographic environment on weather recognition, the model improves its adaptability to different specific environmental scenarios and its recognition accuracy.

[0056] Includes the following steps:

[0057] Several historical panoramic image data from different geographical regions are preprocessed to obtain a preprocessed historical panoramic image dataset. The preprocessing includes image size unification, image enhancement, image grayscale processing, and grayscale value normalization, which transforms the historical panoramic image data into preprocessed historical panoramic image data with uniform size, strong contrast, and grayscale normalization.

[0058] The preprocessed historical panoramic image set is divided into a historical image semantic segmentation training dataset, a historical geographic environment recognition training dataset, and a historical weather recognition training dataset.

[0059] Based on the historical image semantic segmentation training dataset, the GSCNN algorithm is used for optimization training to construct an image semantic segmentation sub-model and obtain the corresponding historical semantic segmentation image dataset.

[0060] By setting the grayscale values ​​of the non-interest regions in the historical semantic segmentation image dataset to 0, the corresponding historical geographic environment segmentation image dataset and historical sky segmentation image dataset are obtained.

[0061] The historical geographic environment segmentation image dataset is added to the historical geographic environment recognition training dataset to obtain the updated historical geographic environment recognition training dataset. The historical sky segmentation image dataset is added to the historical weather recognition training dataset to obtain the updated historical weather recognition training dataset.

[0062] Based on the updated historical geographic environment recognition training dataset, the LFMCNN algorithm is used for optimization training to construct a geographic environment recognition sub-model and obtain several corresponding geographic environment data weight arrays.

[0063] Based on the updated historical weather recognition training dataset, along with corresponding historical geographic environment data and several geographic environment data weight arrays, the ShuffleNet algorithm was used for optimization training to construct a weather recognition sub-model. Geographic environment factors such as time, latitude and longitude, wind speed, wind direction, temperature, humidity, altitude, light intensity, and fog value all have a certain impact on weather recognition. Furthermore, the weights of each geographic environment factor vary depending on the specific environmental scenario. For example, in a rainforest environment, temperature, humidity, and fog value have a disruptive effect on weather recognition because the mean of the above data in the rainforest environment is greater than that in the desert environment, and rainy or foggy days do not significantly change the above data. However, other geographic environment factors can reflect the weather conditions.

[0064] The image semantic segmentation sub-model, the geographic environment recognition sub-model, and the weather recognition sub-model are combined to construct a weather recognition model;

[0065] Acquire real-time panoramic video data and corresponding real-time geographic environment data of the target area;

[0066] Frames are extracted from real-time panoramic video data to obtain continuous frames of real-time panoramic image data.

[0067] Based on real-time panoramic image data of consecutive frames and corresponding real-time geographic environment data, a weather recognition model is used to perform weather recognition to obtain the weather recognition results for the target area, including the following steps:

[0068] Preprocessing and frame skipping are performed on continuous real-time panoramic image data to obtain a preprocessed real-time panoramic image dataset containing several key frames. The preprocessing includes sequential image size unification, image enhancement, image grayscale processing, and grayscale value normalization, converting the real-time panoramic image data into preprocessed real-time panoramic image data with uniform size, strong contrast, and grayscale normalization. Frame skipping avoids analyzing all frames of real-time panoramic image data, thus avoiding resource waste, because weather changes last for a long time and change slowly, and the differences between continuous frames of real-time panoramic image data are not significant, requiring a certain amount of time for changes to occur.

[0069] The preprocessed real-time panoramic image data of the current keyframe is input into the image semantic segmentation sub-model to perform image semantic segmentation and obtain the corresponding real-time semantic segmentation image data.

[0070] By setting the grayscale value of the non-interest regions in the real-time semantic segmentation image data to 0, the corresponding real-time geographic environment segmentation image data and real-time sky segmentation image data are obtained. This removes the interference of non-interest regions on the image, making image recognition more focused on the region of interest and improving the accuracy of image recognition.

[0071] Real-time geographic environment segmentation image data is input into the geographic environment recognition sub-model to perform geographic environment recognition and obtain the corresponding geographic environment category and geographic environment data weight array.

[0072] The real-time sky segmentation image data of the current keyframe, along with the corresponding real-time geographic environment data and geographic environment data weight array, are input into the weather recognition sub-model to perform weather recognition and obtain the weather recognition result of the current keyframe. The real-time geographic environment data includes real-time time, real-time latitude and longitude, real-time wind speed, real-time wind direction, real-time temperature, real-time humidity, real-time altitude, real-time light intensity, and real-time fog value, which are collected simultaneously when acquiring real-time panoramic image data.

[0073] By traversing all key frames of the preprocessed real-time panoramic image data in the preprocessed real-time panoramic image dataset, weather recognition results for several key frames are obtained.

[0074] The weather recognition results of the target area are output as the weather recognition results of the keyframes that belong to the same category and are the most numerous among several keyframes. The weather recognition results include cloudy, foggy, heavy rain, light rain, rainstorm, sunny, typhoon, and snowy days. By selecting the weather recognition results of the same category and the most numerous, the interference of special cases or accidental panoramic image data on the final weather recognition results is avoided, which improves the accuracy and reference value of the weather recognition results of the target area.

[0075] Example 2:

[0076] like Figure 2 As shown, this embodiment provides a visible light visual weather recognition system for implementing a visible light visual weather recognition method. The system includes a visible light camera device and a weather recognition center. The visible light camera device is communicatively connected to the weather recognition center. The visible light camera device is equipped with a visible light camera and a geographic environment data acquisition unit. The weather recognition center is equipped with a weather recognition model and is communicatively connected to an external data server.

[0077] The visible light camera device is used to collect real-time panoramic video data of a target area using a visible light camera, collect real-time geographic environment data of the target area using a geographic environment data acquisition unit, and upload the real-time panoramic video data and corresponding real-time geographic environment data of the target area to a weather recognition center.

[0078] The weather recognition center is used to acquire historical panoramic image data and corresponding historical geographical environment data of different geographical areas sent by external data servers; construct a weather recognition model based on the historical panoramic image data and corresponding historical geographical environment data of different geographical areas; extract frames from real-time panoramic video data to obtain continuous frames of real-time panoramic image data; and perform weather recognition using the weather recognition model based on the continuous frames of real-time panoramic image data and corresponding real-time geographical environment data to obtain the weather recognition result of the target area.

[0079] As a preferred embodiment, the visible light camera device includes a main control unit, a visible light camera, a geographic environment data acquisition unit, and a communication unit. The main control unit is electrically connected to the visible light camera, the geographic environment data acquisition unit, and the communication unit, respectively, and the communication unit is communicatively connected to the weather identification center.

[0080] The main control unit is used to control the normal operation of the visible light camera, the geographic environment data acquisition unit, and the communication unit, and to receive real-time panoramic video data of the target area transmitted by the visible light camera and real-time geographic environment data of the target area transmitted by the geographic environment data acquisition unit.

[0081] A visible light camera is used to collect real-time panoramic video data of the target area and transmit the real-time panoramic video data of the target area to the main control unit.

[0082] The geographic environment data acquisition unit is used to collect real-time geographic environment data of the target area and transmit the real-time geographic environment data of the target area to the main control unit.

[0083] The geographic environment data acquisition unit includes a clock module, latitude and longitude sensors, wind speed sensors, wind direction sensors, temperature sensors, humidity sensors, altitude sensors, light intensity sensors, and fog sensors, used to collect time, latitude and longitude, wind speed, wind direction, temperature, humidity, altitude, light intensity, and fog value of the target area.

[0084] The communication unit is used to upload real-time geographic environment data of the target area and the corresponding real-time geographic environment data to the weather identification center.

[0085] Preferably, the weather recognition center includes a data acquisition unit, a frame capture unit, a model building unit, and a weather recognition unit. The data acquisition unit is connected to a visible light camera and an external data server, and is also connected to the frame capture unit, the weather recognition unit, and the model building unit. The frame capture unit and the model building unit are both connected to the weather recognition unit.

[0086] The data acquisition unit is used to receive real-time geographic environment data of the target area and corresponding real-time geographic environment data uploaded by the visible light camera device, transmit the real-time geographic environment data of the target area to the frame capture unit, and send the real-time geographic environment data of the target area to the weather recognition unit; and to receive several historical panoramic image data and corresponding historical geographic environment data of different geographic environment areas sent by an external data server, and transmit the several historical panoramic image data and corresponding historical geographic environment data of different geographic environment areas to the model building unit.

[0087] The frame capture unit is used to receive real-time geographic environment data of the target area transmitted by the data acquisition unit, capture frames of the real-time panoramic video data to obtain continuous frames of real-time panoramic image data, and transmit the continuous frames of real-time panoramic image data to the weather recognition unit.

[0088] The model building unit is used to receive several historical panoramic image data and corresponding historical geographical environment data of different geographical environment areas transmitted by the data acquisition unit, construct a weather recognition model based on the several historical panoramic image data and corresponding historical geographical environment data of different geographical environment areas, and transmit the weather recognition model to the weather recognition unit.

[0089] The weather recognition unit receives the weather recognition model transmitted by the model building unit, the real-time panoramic image data of consecutive frames transmitted by the frame capture unit, and the real-time geographic environment data of the target area transmitted by the data acquisition unit. Based on the real-time panoramic image data of consecutive frames and the corresponding real-time geographic environment data, the weather recognition model is used to perform weather recognition to obtain the weather recognition result of the target area.

[0090] This invention provides a visible light visual weather recognition method and system. By using panoramic video data acquired through visible light visual technology and image recognition technology, a weather recognition model is constructed for automated, real-time weather recognition. This avoids reliance on manual analysis and judgment, improving the accuracy and efficiency of weather recognition and ensuring its timeliness. Furthermore, by using geographic environmental data acquired through sensor technology, the impact of geographic environmental factors on weather recognition is considered, improving the adaptability and accuracy of the weather recognition method and system to special outdoor environments. This also avoids the need to build weather stations in specific areas of the field, reducing the cost of weather recognition.

[0091] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the teachings of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A visible light visual weather recognition method, characterized in that: Includes the following steps: A weather recognition model is constructed based on several historical panoramic image data and corresponding historical geographical environment data of different geographical regions. The weather recognition model includes an image semantic segmentation sub-model, a geographic environment recognition sub-model, and a weather recognition sub-model. The image semantic segmentation sub-model is constructed based on the GSCNN algorithm and includes a first input layer, a rule flow module, a shape flow module, a fusion module, and a first output layer. The first input layer is connected to the rule flow module and the shape flow module, respectively. The rule flow module and the shape flow module are both connected to the fusion module. The fusion module is connected to the first output layer. The first output layer is connected to the geographic environment recognition sub-model and the weather recognition sub-model, respectively. The geographic environment recognition sub-model is built based on the LFMCNN algorithm and includes a second input layer, a first convolutional module, a downsampling module, a second convolutional module, a third convolutional module, a fully connected layer, and a second output layer. The second input layer is connected to the first output layer, the first convolutional module, and the downsampling module of the image semantic segmentation sub-model. The first convolutional module and the downsampling module are both connected to the second convolutional module. The second convolutional module, the third convolutional module, the fully connected layer, and the second output layer are connected in sequence. The second output layer is connected to the weather recognition sub-model. The weather recognition sub-model is built based on the ShuffleNet algorithm and includes a third input layer, a first group convolutional layer, a feature fusion layer, a Shuffle layer, a second group convolutional layer, and a third output layer. The third input layer is connected to the first output layer and the first group convolutional layer of the image semantic segmentation sub-model, respectively. The first group convolutional layer, the feature fusion layer, the Shuffle layer, the second group convolutional layer, and the third output layer are connected in sequence. The feature fusion layer is connected to the second output layer of the geographic environment recognition sub-model. Acquire real-time panoramic video data and corresponding real-time geographic environment data of the target area; Frames are extracted from real-time panoramic video data to obtain continuous frames of real-time panoramic image data. Preprocessing and frame skipping filtering are performed on continuous real-time panoramic image data to obtain a preprocessed real-time panoramic image dataset containing several key frames. The preprocessed real-time panoramic image data of the current keyframe is input into the image semantic segmentation sub-model to obtain the corresponding real-time semantic segmentation image data. By setting the grayscale value of the non-interest region of the real-time semantic segmentation image data to 0, the corresponding real-time geographic environment segmentation image data and real-time sky segmentation image data are obtained. Real-time geographic environment segmentation image data is input into the geographic environment recognition sub-model to obtain the corresponding geographic environment category and geographic environment data weight array. The real-time sky segmentation image data of the current keyframe, along with the corresponding real-time geographic environment data and geographic environment data weight array, are input into the weather recognition sub-model to obtain the weather recognition result of the current keyframe. By traversing the preprocessed real-time panoramic image dataset containing several keyframes, weather recognition results for several keyframes are obtained. The weather recognition results of the keyframes that belong to the same category and have the largest number of keyframes are output as the weather recognition results for the target area.

2. The visible light visual weather recognition method according to claim 1, characterized in that: Based on several historical panoramic image data and corresponding historical geographical environment data of different geographical regions, a weather recognition model is constructed, including the following steps: Several historical panoramic image data from different geographical regions were preprocessed to obtain a preprocessed historical panoramic image dataset. The preprocessed historical panoramic image set is divided into a historical image semantic segmentation training dataset, a historical geographic environment recognition training dataset, and a historical weather recognition training dataset. Based on the historical image semantic segmentation training dataset, the GSCNN algorithm is used for optimization training to construct an image semantic segmentation sub-model and obtain the corresponding historical semantic segmentation image dataset. By setting the grayscale values ​​of the non-interest regions in the historical semantic segmentation image dataset to 0, the corresponding historical geographic environment segmentation image dataset and historical sky segmentation image dataset are obtained. The historical geographic environment segmentation image dataset is added to the historical geographic environment recognition training dataset to obtain the updated historical geographic environment recognition training dataset. The historical sky segmentation image dataset is added to the historical weather recognition training dataset to obtain the updated historical weather recognition training dataset. Based on the updated historical geographic environment recognition training dataset, the LFMCNN algorithm is used for optimization training to construct a geographic environment recognition sub-model and obtain several corresponding geographic environment data weight arrays. Based on the updated historical weather recognition training dataset, as well as several historical geographical environment data and several geographical environment data weight arrays, the ShuffleNet algorithm is used for optimization training to construct a weather recognition sub-model. The image semantic segmentation sub-model, the geographic environment recognition sub-model, and the weather recognition sub-model are combined to construct a weather recognition model.

3. The visible light visual weather recognition method according to claim 1, characterized in that: The types of historical / real-time geographic environment data include time, latitude and longitude, wind speed, wind direction, temperature, humidity, altitude, light intensity, and fog value.

4. The visible light visual weather recognition method according to claim 1, characterized in that: The preprocessing includes sequential image size unification, image enhancement, image grayscale processing, and grayscale value normalization.

5. A visible light visual weather recognition system, used to implement the visible light visual weather recognition method as described in any one of claims 1-4, characterized in that: The system includes a visible light camera device and a weather recognition center. The visible light camera device is communicatively connected to the weather recognition center and is equipped with a visible light camera and a geographic environment data acquisition unit. The weather recognition center is equipped with a weather recognition model and is communicatively connected to an external data server. The visible light camera device is used to collect real-time panoramic video data of a target area using a visible light camera, collect real-time geographic environment data of the target area using a geographic environment data acquisition unit, and upload the real-time panoramic video data and corresponding real-time geographic environment data of the target area to a weather recognition center. The visible light camera device includes a main control unit, a visible light camera, a geographic environment data acquisition unit, and a communication unit. The main control unit is electrically connected to the visible light camera, the geographic environment data acquisition unit, and the communication unit, respectively. The communication unit is communicatively connected to the weather identification center. The weather recognition center is used to acquire historical panoramic image data and corresponding historical geographical environment data of different geographical areas sent by external data servers; construct a weather recognition model based on the historical panoramic image data and corresponding historical geographical environment data of different geographical areas; extract frames from real-time panoramic video data to obtain continuous frames of real-time panoramic image data; and use the weather recognition model to perform weather recognition based on the continuous frames of real-time panoramic image data and corresponding real-time geographical environment data to obtain the weather recognition result of the target area. The weather recognition center includes a data acquisition unit, a frame capture unit, a model building unit, and a weather recognition unit. The data acquisition unit is connected to a visible light camera and an external data server, and is also connected to the frame capture unit, the weather recognition unit, and the model building unit. The frame capture unit and the model building unit are both connected to the weather recognition unit.

Images