Squid fishing information acquisition and data cloud storage method
By using flexible RFID tags and YOLO neural network identification models, combined with GPS and cloud databases, squid catch information is automatically recorded and uploaded. This solves the problem of automatic collection and data acquisition of squid catch information, realizes automated storage of squid catch data, and ensures the real-time and reliability of squid catch data, thereby enhancing the transparency and traceability of squid catch information.
Patent Information
- Application Number
- CN202511135936.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-14
- Publication Date
- 2025-11-28
AI Technical Summary
Current technologies lack real-time monitoring of squid catch information collection, and data silos are severe in the processing and distribution process, resulting in insufficient transparency and traceability of the supply chain. Furthermore, the efficiency of binding information collection entities with storage carriers is low, requiring significant environmental modifications.
Flexible RFID tags are used to bind fish catch identification codes. Combined with a fish catch grade recognition model based on YOLO neural network, the fish catch grade is automatically identified. The data is persistently stored in a cloud database via network connection. The Modbus RTU protocol and GPS are used to obtain the geographical location, enabling automatic recording and uploading.
It enables the automatic collection and cloud storage of squid catch information. Through automatic recording and uploading, it simplifies the operation, improves the data collection, real-time performance and reliability, adapts to multiple scenarios, and enhances the transparency and traceability of the squid supply chain.
Smart Images
Figure CN121031633A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of information collection and data storage, in particular to a squid catch information collection and data cloud storage method. BACKGROUND
[0002] Catch refers to aquatic biological resources obtained from water through fishing activities, mainly including fish, shellfish, crustaceans (such as shrimp, crab), cephalopods (such as squid), and algae, etc. Squid is a high-quality marine product with high protein and low fat, and its international trade is increasing. However, there are significant pain points in the current industry chain: lack of real-time monitoring in the fishing link, and serious data islandization in the processing and circulation link. In order to enhance the transparency and traceability of the supply chain, a squid catch information collection and data persistent storage method is needed.
[0003] Currently, catch is not recorded in detail during production, transportation, and landing, which is not conducive to the management and traceability of the whole link. In most existing technical solutions, the information collection subject and the storage carrier are bound to have a manual collection step, which is low in efficiency; and most existing technical solutions require substantial modification of the operating environment, and do not consider the problem of multi-scene adaptability. Therefore, in order to enhance the transparency and traceability of the squid supply chain, a squid catch information collection and data cloud storage method is needed, which is used to automatically record the time, location, grade, etc. of squid catch production, transportation, and landing, and upload to the cloud database for persistent storage. SUMMARY
[0004] In view of the defects in the prior art, the purpose of the present application is to provide a squid catch information collection and data cloud storage method, which is used to automatically record the time, location, grade, etc. of squid catch production, transportation, and landing, and upload to the cloud database for persistent storage.
[0005] To solve the above problems, the technical solution of the present application is as follows:
[0006] A squid catch information collection and data cloud storage method, comprising the following steps:
[0007] Obtaining a catch identification code;
[0008] Using a catch grade recognition model based on a YOLO neural network to automatically recognize the catch grade;
[0009] After obtaining the RFID identification code and the catch grade of the catch, all data are combined into a single data record and temporarily stored in a local data list;
[0010] Connecting the cloud database through a network to batch upload the temporarily stored data in the local data list, and if the upload fails, the failed data record is temporarily stored in the local database.
[0011] Preferably, the step of acquiring the catch identification code specifically includes: placing the squid catch provided with the RFID tag within the identification range of the host computer, the RFID reader built-in the host computer reads the unique identification code carried by the tag, and the identification code is transmitted to the on-board program through the serial port.
[0012] Preferably, the step of automatically identifying the catch grade using the catch grade identification model based on the YOLO neural network specifically includes:
[0013] First, initialization configuration is performed, including loading the YOLO model, reading the Chinese label configuration, and setting the video input source;
[0014] After initialization is completed, a FrameGrabber thread is created to asynchronously process the video stream, which is responsible for continuously obtaining video frames from the input source and putting them into a fixed-size queue, obtaining video frames from the FrameGrabber queue, and pre-processing each frame;
[0015] The pre-processed image is sent to the YOLO model for inference, which detects the catch in the image and classifies it according to different labels on its packaging, and each detection result contains the bounding box position, catch grade, and confidence information.
[0016] Preferably, the catch grade identification model of the YOLO neural network specifically includes:
[0017] Image acquisition and preprocessing: real-time video stream or static images containing catches are obtained through an image acquisition device, and standardization preprocessing is performed on the images, including size normalization and pixel value standardization;
[0018] Attention-enhanced feature extraction: the pre-processed image is input into a deep convolutional neural network model, the backbone network of which is based on the core structure of YOLOv8 and is responsible for extracting hierarchical features of the image, to improve the recognition accuracy of YOLOv8 in low-light and possibly stained environments on the ship, a parameter-free attention mechanism module is integrated into the backbone network; the backbone network generates multi-scale feature maps from the input image through stacked convolutional layers and C2f modules; the parameter-free attention mechanism module is deployed after at least one C2f module;
[0019] Classification prediction: the attention-enhanced feature map is passed to the classification head of the model, which converts the high-dimensional feature map into a one-dimensional feature vector through a global average pooling operation; then, the feature vector passes through one or more fully connected layers and is finally processed by a Softmax activation function, outputting a probability distribution vector; each dimension of the vector corresponds to a predefined catch grade, and its value represents the confidence probability that the catch in the image belongs to that grade;
[0020] Result output and stability processing: the system selects the grade with the highest probability value as the prediction result of the single frame image.
[0021] Preferably, after obtaining the RFID identification code of the catch and the catch grade, the step of combining all data into a single data record and temporarily storing it in the local data list comprises: after obtaining the RFID identification code of the catch and the catch grade, the upper computer connected to the Internet communicates with the COM serial port connected to the GPS device through the serial module of python using the Modbus RTU protocol, sends a read dimension Modbus instruction, and obtains the current geographic position of the upper computer as the geographic position bound to the catch; the RFID identification code of the catch, the catch grade, the geographic position, the current time and the ship number are combined into a single data record and temporarily stored in the local data list.
[0022] Preferably, the step of connecting to the cloud database through the network, uploading the temporarily stored data in the local data list in batches, and if the uploading fails, temporarily storing the failed data records in the local database comprises: connecting the database using the pymysql module of python and uploading all data records for cloud persistent storage, and when there is no network and the uploading fails, inserting the failed data into the local database for temporary storage; after a round of production operation is completed, the temporarily stored failed data is exported as a table, and other network-enabled electronic devices are used to log in to the data uploading subsystem to re-upload the data.
[0023] Compared with the prior art, the present application has the following advantages:
[0024] 1. The present application uses a flexible RFID tag to bind the identification code with the catch, uses the RFID code as the unique identifier of a single package of catch, and associates it with other information, realizing reliable traceability of one thing to one code;
[0025] 2. The present application stores the catch data persistently in the cloud, which is beneficial to the management and traceability of the whole chain of catch;
[0026] 3. The present application automatically records the time, location, grade and other information of the squid catch production, transportation and landing, effectively improving the efficiency and simplifying the operation difficulty. BRIEF DESCRIPTION OF DRAWINGS
[0027] Other features, objects, and advantages of the application will become more apparent from the following detailed description when read in conjunction with the accompanying drawings:
[0028] Figure 1 A flow chart of the method for collecting squid catch information and storing data in the cloud according to the application;
[0029] Figure 2 A detailed flow chart of the method for collecting squid catch information and storing data in the cloud according to the application;
[0030] Figure 3 A framework diagram for catch grade identification. DETAILED DESCRIPTION
[0031] The application will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the application, but do not limit the application in any form. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the application. These are within the scope of protection of the application.
[0032] Specifically, the application provides a method for collecting squid catch information and storing data in the cloud, as shown in Figure 1 and Figure 2 The method comprises the following steps: using an upper computer with a multi-expansion interface and an on-board program as the main carrier of the system, and a cloud database as the end node, to realize the method for collecting squid catch information and storing data in the cloud proposed by the application, wherein the upper computer comprises a host computer, an RFID reader / writer, a network module, a GPS positioning module, and an expansion serial port.
[0033] S1: Obtain a catch identification code;
[0034] Specifically, the squid catch provided with a flexible RFID tag is placed within the identification range of the upper computer, and the RFID reader / writer built-in the upper computer reads the unique identification code carried by the tag, and transmits the identification code to the on-board program through the serial port. The RFID technology can ensure the uniqueness and integrity of each catch individual, and the use of RFID technology realizes the binding of the catch and the identification code and the information collection, which is suitable for automatic reading of multiple packages of catches, is not affected by stains or light, has high reading efficiency, and is convenient to obtain the catch identification.
[0035] S2: Use a catch grade identification model based on a YOLO neural network to automatically identify the catch grade;
[0036] The application designs a catch grade identification model based on a YOLO neural network to realize full-automatic acquisition of the catch grade.
[0037] Specifically, as shown in Figure 3As shown, first, the initialization configuration is performed, including loading the YOLO model, reading the Chinese label configuration, and setting the video input source. The application supports three input sources: local camera, video file, and RTSP network camera stream. After initialization, a FrameGrabber thread is created to asynchronously process the video stream. This thread is responsible for continuously obtaining video frames from the input source and placing them into a fixed-size queue to ensure real-time performance and memory efficiency. Video frames are obtained from the FrameGrabber queue, and each frame is preprocessed, including adjusting the image size and format conversion. The preprocessed image is sent to the YOLO model for inference. The model detects fish catches in the image and classifies them according to different labels on their packaging. Each detection result contains the bounding box position, fish catch level, and confidence information. The detection result is post-processed, and Chinese annotations are added to the image using the PIL library, including class names and confidence. The processed image is adjusted to the appropriate display size, then displayed in real time through the OpenCV window, and the fish catch level label is output. During the entire process, the module continuously monitors the availability of the video source. If it detects that the video stream has ended or an error has occurred, it will release resources and exit the program. This flow design fully considers real-time performance, reliability, and user experience. Through multi-threading processing, it ensures the smoothness of the video stream, and through asynchronous processing, it avoids interface lag. At the same time, it provides a flexible error handling mechanism. Modular design also makes it easy to maintain and extend, allowing new features to be added or existing features to be optimized as needed.
[0038] The architecture of the fish catch level recognition model of the YOLO neural network specifically includes:
[0039] Image acquisition and preprocessing: real-time video streams or static images containing fish catches are obtained through image acquisition devices (such as industrial cameras), and standard preprocessing is performed on the images, including size normalization and pixel value standardization.
[0040] Attention-enhanced feature extraction: the preprocessed image is input into a deep convolutional neural network model, and the backbone network of the model is based on the core structure of YOLOv8, responsible for extracting hierarchical features of the image. To improve the recognition accuracy of YOLOv8 in low-light and possibly stained environments on the ship, a parameter-free attention mechanism module is integrated at several key nodes within the backbone network.
[0041] The backbone network generates multi-scale feature maps from the input image through stacked convolutional layers and C2f modules. The parameter-free attention mechanism module is strategically deployed after at least one C2f module. According to the principles of neuroscience, this attention module evaluates the importance of each neuron in the feature map by defining an energy function for it.
[0042] Specifically, for a target neuron t in the feature map, its energy function is The energy function value is calculated by the following closed-form solution:
[0043]
[0044] where t is the eigenvalue of the target neuron, μ and σ 2 are the mean and variance of all neurons in the feature channel, respectively, and the calculation formula is: and where x i is the eigenvalue of the i-th neuron in the channel, and M is the total number of neurons on the channel. λ is a hyperparameter as a regularization coefficient.
[0045] The smaller the energy function value of the target neuron t, the higher the discrimination of the target neuron t from other neurons around it, and the higher the importance of the target neuron t. Therefore, the importance (i.e., attention weight) of each neuron is defined as By calculating the weight for each neuron, the module can generate a three-dimensional attention map, and the weight map is used to recalibrate the feature map, thereby significantly enhancing the expression of features with high discrimination for the fishing grade classification task while suppressing the interference of background noise and irrelevant visual information without increasing the computational burden and parameters of the model.
[0046] Classification prediction: the attention-enhanced feature map is passed to the classification head of the model, which converts the high-dimensional feature map into a one-dimensional feature vector through global average pooling operation; then, the feature vector is processed by one or more fully connected layers and finally by a Softmax activation function, outputting a probability distribution vector, each dimension of which corresponds to a predefined fishing grade, and the value represents the confidence probability that the catch in the image belongs to the grade.
[0047] Result output and stability processing: the system selects the grade with the highest probability value as the prediction result of the single frame image.
[0048] For video stream processing, the present application also includes a time series stability module that caches the prediction results of the last N frames and only outputs a specific grade as the final and reliable classification result when the prediction frequency of the grade exceeds a preset stability threshold within the cache window. This effectively avoids the problem of frequent prediction result jumps caused by small inter-frame disturbances, enhancing the stability and reliability of the system in continuous fishing information entry operations.
[0049] S3: After obtaining the RFID identification code of the catch and the fishing grade, all data are combined into a single data record and temporarily stored in the local data list;
[0050] Specifically, after obtaining the RFID identification code of the catch and the catch level, the upper computer connected to the Internet communicates with the COM serial port connected to the GPS device through the serial module of python using the Modbus RTU protocol, sends the read dimensional Modbus instruction, and obtains the current geographic position of the upper computer as the geographic position bound to the catch. The RFID identification code of the catch, the catch level, the geographic position, the current time and the ship number are combined into a single data record and temporarily stored in the local data list.
[0051] S4: Connect to the cloud database through the network, batch upload the temporarily stored data in the local data list, and if the upload fails, temporarily store the failed data records in the local database.
[0052] Specifically, the pymysql module of python is used to connect the database and upload all data records for cloud persistent storage, and when there is no network to upload, the failed data is inserted into the local database for temporary storage. After a round of production operation is completed, the temporarily stored failed data is exported as a table, and other network-enabled electronic devices are used to log in to the data upload subsystem to re-upload the data.
[0053] The specific embodiments of the application are described above. It should be understood that the application is not limited to the specific implementation described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essential content of the application. The embodiments of the present application and the features in the embodiments can be arbitrarily combined with each other without conflict.
Claims
1. A method for collecting squid catch information and storing data in the cloud, characterized in that, The method includes the following steps: Obtain the catch identification code; The catch grade is automatically identified using a YOLO neural network-based catch grade recognition model. After obtaining the RFID tag and catch grade of the catch, all data is combined into a single data record and temporarily stored in the local data list. By connecting to a cloud database via the network, temporary data from the local data list can be uploaded in batches. If the upload fails, the failed data record is temporarily stored in the local database.
2. The method for collecting squid catch information and storing data in the cloud according to claim 1, characterized in that, The steps for obtaining the catch identification code specifically include: placing the squid catch with RFID tags within the identification range of the host computer, the RFID reader built into the host computer reading the unique identification code carried by the tag, and transmitting the identification code to the onboard program via serial port.
3. The method for collecting squid catch information and storing data in the cloud according to claim 1, characterized in that, The step of automatically identifying the catch grade using a YOLO neural network-based catch grade recognition model specifically includes: First, perform initial configuration, including loading the YOLO model, reading the Chinese label configuration, and setting the video input source; After initialization, a FrameGrabber thread is created to asynchronously process the video stream. This thread is responsible for continuously obtaining video frames from the input source and putting them into a queue of fixed size. Video frames are obtained from the FrameGrabber queue, and each frame is preprocessed. The preprocessed image is fed into the YOLO model for inference. The model detects the catch in the image and classifies it according to different markings on its packaging. Each detection result includes bounding box location, catch grade, and confidence information.
4. The method for collecting squid catch information and storing data in the cloud according to claim 1, characterized in that, The YOLO neural network-based fish catch rating model specifically includes: Image acquisition and preprocessing: real-time video streams or still images containing the catch are acquired through an image acquisition device, and the images are standardized preprocessed, including size normalization and pixel value normalization. Attention-enhanced feature extraction: The preprocessed image is input into a deep convolutional neural network model. The backbone network of this model is based on the core structure of YOLOv8 and is responsible for extracting hierarchical features of the image. To improve the recognition accuracy of YOLOv8 in low-light and potentially stained environments on ships, a parameterless attention mechanism module is integrated into the backbone network. The backbone network generates multi-scale feature maps from the input image through stacked convolutional layers and C2f modules. The parameterless attention mechanism module is deployed after at least one C2f module. Classification prediction: The attention-enhanced feature map is passed to the model's classification head, which converts the high-dimensional feature map into a one-dimensional feature vector through a global average pooling operation. Subsequently, the feature vector is passed through one or more fully connected layers and finally processed by a Softmax activation function to output a probability distribution vector. Each dimension of this vector corresponds to a predefined catch level, and its value represents the confidence probability that the catch in the image belongs to that level. Output Results and Stability Processing: The system selects the level with the highest probability value as the prediction result for a single frame image.
5. The method for collecting squid catch information and storing data in the cloud according to claim 1, characterized in that, The step of combining all data into a single data record and temporarily storing it in a local data list after obtaining the RFID tag and catch grade of the catch specifically includes: after obtaining the RFID tag and catch grade of the catch, the host computer connected to the Internet communicates with the COM serial port of the GPS device using the Modbus RTU protocol through the serial module of Python, and sends a Modbus command to read the latitude and longitude, thereby obtaining the current geographical location of the host computer as the geographical location bound to the catch; the RFID tag, catch grade, geographical location, current time and boat number of the catch are combined into a single data record and temporarily stored in a local data list.
6. The method for collecting squid catch information and storing data in the cloud according to claim 1, characterized in that, The steps of connecting to a cloud database via network to batch upload temporary data from a local data list, and temporarily storing the failed data records in the local database if the upload fails, specifically include: using Python's pymysql module to connect to the database and upload all data records for persistent cloud storage; when uploading fails due to lack of network, inserting the failed data into the local database for temporary storage; after a round of production operations is completed, exporting the temporarily stored failed data into a table, and using other network-enabled electronic devices to log in to the data upload subsystem to re-upload the data.