Agricultural product pesticide residue detection system based on artificial intelligence

Abstract

The invention discloses an agricultural product pesticide residue detection system based on artificial intelligence, and relates to the technical field of agricultural product detection. The system comprises a data acquisition module, a preprocessing module, a standard database module, a result output module and a communication module. The data acquisition module acquires spectral image data of agricultural products by using a hyperspectral imaging technology; the preprocessing module is used for denoising, enhancing and correcting an image and extracting features through various algorithms, and analyzing pesticide residues by using a deep learning model; the standard database module stores related information of pesticides, standard limits and crops; the result output module is used for outputting qualification, disqualification or to-be-rechecked judgment according to a standard comparison analysis result, and generating a detailed detection report; and the communication module realizes data transmission of each module and interaction with a remote server. The method solves the problems of complex detection, long time, high cost, destructive treatment of samples and the like of a traditional detection method, has the characteristics of rapidness, no damage and accuracy, can meet different standard requirements, and is of great significance for guaranteeing the quality and safety of agricultural products.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural product detection, and more specifically, to an agricultural product pesticide residue detection system based on artificial intelligence. Background Art

[0002] As living standards improve, demands for the quality and safety of agricultural products are becoming increasingly stringent. Pesticide residues pose a serious threat to human health, making accurate and efficient detection of pesticide residues in agricultural products crucial. Traditional pesticide residue detection methods, such as gas chromatography and liquid chromatography, while highly accurate, suffer from complex, time-consuming, and costly processes, as well as the need for destructive sample processing. Furthermore, these methods require high levels of expertise from testers, making large-scale, rapid on-site testing difficult.

[0003] Agricultural product pesticide residue testing still faces challenges such as difficulty collecting test data, inefficient data processing, and an inability to quickly and accurately determine whether test results meet relevant standards. Furthermore, due to varying pesticide residue standards across countries and regions, existing testing systems struggle to meet diverse standards. Therefore, developing a rapid, non-destructive, accurate, and adaptable agricultural product pesticide residue testing system is of great practical significance. Summary of the Invention

[0004] The purpose of the present invention is to provide an artificial intelligence-based agricultural product pesticide residue detection system to solve the problems existing in existing agricultural product pesticide residue detection methods, such as complex detection process, long detection time, high cost, need for destructive sample treatment, and difficulty in detection data collection and processing.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions: an artificial intelligence-based agricultural product pesticide residue detection system, a data acquisition module, including a hyperspectral imaging unit and a data transmission unit; obtaining spectral image data of agricultural products through hyperspectral imaging technology and transmitting the data to a data processing module;

[0006] The data preprocessing module, including the image preprocessing subunit, feature extraction subunit and pesticide residue analysis subunit, preprocesses and extracts features from the collected spectral image data; and analyzes the types and amounts of pesticide residues on the processed data by building a deep learning model;

[0007] The standard database module includes a pesticide variety database, a standard limit database, and a crop name database, which is used to store relevant information on pesticide varieties, standard limits, and crop names;

[0008] The result output module includes a data comparison subunit that uses multi-threaded parallel processing technology and establishes standard priority rules, and a data result output subunit with the function of generating detailed test reports; it is used to compare the pesticide residue analysis results of the pesticide residue analysis subunit with the standards in the standard database module and output the test results;

[0009] The communication module is used to realize data transmission between modules and connection with the remote server.

[0010] The present invention is further configured as follows: a hyperspectral imaging unit in a data acquisition module adopts a linear array CCD detector, a hyperspectral camera with a spectral range of 400-1000nm, a spectral resolution of 2.5nm and a spatial resolution of 0.1mm; the hyperspectral imaging unit is provided with an adjustable stage, the distance between the camera and the stage is maintained at 30-50cm, the hyperspectral camera is controlled by an electric slide to scan at a speed of 0.1-1m / s, and is provided with a halogen lamp and LED combination light source; the data transmission unit adopts a Gigabit Ethernet interface based on the TCP / IP protocol, is equipped with a data cache module and uses a compression algorithm; the compression ratio of the data transmission unit is between 1:10 and 1:20.

[0011] The present invention is further configured as follows: a data preprocessing module uses a three-dimensional block matching algorithm for denoising, a deep learning-based Retinex enhancement network to enhance the image, and a polynomial correction method to perform geometric correction in the image preprocessing subunit; an attention mechanism module is introduced and a feature extraction subunit of a convolutional neural network model trained with a transfer learning strategy is adopted; the geometric correction of the image preprocessing subunit uses at least 4 ground control points, combined with camera intrinsic and extrinsic parameter information, to control the image geometric distortion error within 0.5 pixels.

[0012] The present invention is further configured as follows: the feature extraction subunit constructs an improved ResNet-50 convolutional neural network model, based on the ImageNet pre-trained model, fine-tunes it on a dataset containing 100,000 annotated hyperspectral images, uses the Adam optimizer, sets the initial learning rate to 0.001, decays by 0.1 every 10 epochs, and trains for 50 epochs.

[0013] The present invention is further configured as follows: the pesticide residue analysis subunit adopts a support vector regression model, uses 20,000 sets of sample data for training, optimizes the penalty parameter C and the RBF kernel function width γ through cross-validation, and the model prediction mean square error is less than 0.05.

[0014] The present invention is further configured as follows: the crop name database records in detail the biological classification information of the crop kingdom, phylum, class, order, family, genus, and species; the standard limit database uses crawler technology to monitor the official agency websites of various countries or regions in real time, and automatically downloads when the standards are updated and is stored after manual review and verification.

[0015] The present invention is further configured as follows: a data comparison subunit automatically determines the standard priority based on information such as the sales destination and production place of agricultural products; the test report generated by the result output subunit supports text report template generation, Echarts chart interactive display, and local SQLite and cloud RDS data management.

[0016] The present invention is further configured such that: the result output module is provided with a judgment rule engine, which outputs judgment results including three states: qualified, unqualified and to be re-inspected.

[0017] The present invention is further configured such that: the communication module adopts the TCP / IP protocol, uses WebSocket to transmit real-time data, the data is encrypted via SSL / TLS, and supports interaction with distributed remote servers.

[0018] The present invention is further configured as follows: the remote server adopts a distributed architecture and deploys multiple server nodes to achieve load balancing and high availability; after the detection terminal establishes a connection with the server through the communication module, it first performs identity authentication, and after the authentication is passed, the detection data is packaged in a prescribed format and uploaded to the server; after the server receives the data, it performs data storage, backup and preliminary analysis and processing; at the same time, the server can push the latest model parameters, standard database update information, and system upgrade patches to the detection terminal; after receiving the information pushed by the server, the detection terminal automatically performs corresponding update operations to ensure that the system is always kept up to date.

[0019] The present invention is further configured as follows: a terminal adaptation module and a cross-platform data service layer are also provided. The terminal adaptation module is responsible for identifying the device type, operating system version, and screen resolution of the mobile phone and computer terminals, and formatting the test result data according to the characteristics of different terminals; the cross-platform data service layer adopts a RESTful API architecture as a data interaction interface between the detection system and the mobile phone and computer applications to realize unified management and distribution of data; the test results can be viewed on the mobile phone and computer terminals.

[0020] In summary, the present invention has the following beneficial effects:

[0021] Efficient and rapid testing: Using hyperspectral imaging technology combined with advanced data processing algorithms, the system significantly reduces testing time compared to traditional methods, enabling rapid testing and meeting the needs of large-scale agricultural product testing. For example, traditional gas chromatography testing of a batch of agricultural products may take several hours or even longer, while this system can complete multiple batches of testing in a short period of time.

[0022] Non-destructive testing: avoids destructive treatment of agricultural products, ensures the integrity and marketability of agricultural products, and reduces the losses caused by testing to agricultural products.

[0023] Accurate Detection: By building deep learning models and optimizing algorithms, such as the improved ResNet-50 convolutional neural network model and support vector regression model, the accuracy of pesticide residue detection is improved and the error is reduced. The model's prediction mean square error is less than 0.05, accurately determining the type and content of pesticide residues.

[0024] Strong standard adaptability: The standard database module uses crawler technology to update standards in real time. The data comparison sub-unit can automatically determine the standard priority based on information such as the sales destination and production location of agricultural products, and can adapt to the diverse standard requirements of different countries and regions.

[0025] Convenient Data Management and Interaction: Equipped with comprehensive data management capabilities, the result output module supports test report generation and data display in multiple formats, making it easy for users to view and analyze data. Furthermore, through the terminal adaptation module and cross-platform data service layer, convenient viewing and interaction are achieved on mobile phones and computers, allowing users to access test results anytime, anywhere.

[0026] System self-update: The communication module and remote server work together to enable the detection terminal to automatically update model parameters, standard database and system upgrade patches, ensuring that the system is always kept up to date and improving detection performance and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a functional module diagram of a system for detecting pesticide residues in agricultural products according to an embodiment of the present invention;

[0028] Figure 2 This is a functional diagram of a data acquisition module in an embodiment of the present invention;

[0029] Figure 3 This is a functional diagram of a data preprocessing module in an embodiment of the present invention;

[0030] Figure 4 This is a functional diagram of a standard database module in an embodiment of the present invention;

[0031] Figure 5 This is a functional diagram of a result output module in an embodiment of the present invention;

[0032] Figure 6 This is a functional diagram of the communication module, terminal adaptation module, and cross-platform data service layer of the agricultural product pesticide residue detection system according to an embodiment of the present invention;

[0033] Figure 7 This is a flow chart of a system for detecting pesticide residues in agricultural products according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following is combined with Figure 1-7 The present invention is described in further detail.

[0035] Example: An artificial intelligence-based agricultural product pesticide residue detection system, such as Figure 1-Figure 7 As shown, the data acquisition module includes a hyperspectral imaging unit and a data transmission unit; the spectral image data of agricultural products is obtained through hyperspectral imaging technology, and the data is transmitted to the data processing module;

[0036] The data preprocessing module, including the image preprocessing subunit, feature extraction subunit and pesticide residue analysis subunit, preprocesses and extracts features from the collected spectral image data; and analyzes the types and amounts of pesticide residues on the processed data by building a deep learning model;

[0037] The standard database module includes a pesticide variety database, a standard limit database, and a crop name database, which is used to store relevant information on pesticide varieties, standard limits, and crop names;

[0038] The result output module includes a data comparison subunit that uses multi-threaded parallel processing technology and establishes standard priority rules, and a data result output subunit with the function of generating detailed test reports; it is used to compare the pesticide residue analysis results of the pesticide residue analysis subunit with the standards in the standard database module and output the test results;

[0039] The communication module is used to realize data transmission between modules and connection with the remote server.

[0040] A hyperspectral camera equipped with a linear array CCD detector is used, with a spectral range of 400-1000nm, a spectral resolution of 2.5nm, and a spatial resolution of 0.1mm. To accommodate the testing of agricultural products of varying sizes, an adjustable stage is provided, ranging in size from 30cm×30cm to 60cm×60cm. During testing, the distance between the camera and the stage is maintained at 30-50cm. The camera is controlled by an electric slide for uniform scanning, with a scanning speed adjustable within the range of 0.1-1m / s to ensure the acquisition of complete and clear spectral image data. Furthermore, the built-in light source system utilizes a combination of halogen lamps and LEDs, which can adjust the light intensity and spectral distribution according to the characteristics of the agricultural products to ensure the accuracy of the collected data.

[0041] Data transmission utilizes a Gigabit Ethernet interface based on the TCP / IP protocol and is equipped with a 1GB data cache module to prevent data loss due to differences in data transmission speeds. During transmission, data is compressed using the JPEG2000 compression algorithm, achieving a compression ratio of 1:10-1:20, improving transmission efficiency while ensuring data integrity and recoverability.

[0042] The image preprocessing subunit uses a three-dimensional block matching (BM3D) algorithm for denoising, targeting the spectral-spatial characteristics of hyperspectral images to effectively remove noise while preserving image detail. Image enhancement utilizes a deep learning-based Retinex enhancement network. By learning from a large amount of image data, it adaptively adjusts image brightness and contrast, highlighting spectral features related to pesticide residues. Geometric correction employs a polynomial correction method, utilizing at least four ground control points and combining the intrinsic and extrinsic parameters of the hyperspectral camera to control image geometric distortion errors to within 0.5 pixels.

[0043] The feature extraction subunit constructs a modified ResNet-50 convolutional neural network model and introduces a CBAM (Central Attention Mechanism) into the network structure to enhance the network's sensitivity to pesticide residue features. During the model training phase, a transfer learning strategy was employed, using a pre-trained ImageNet model as a foundation and fine-tuning on a self-built dataset of 100,000 annotated hyperspectral images. During training, the Adam optimizer was used, with an initial learning rate of 0.001 and a decay of 0.1 every 10 epochs. After 50 epochs of training, the model achieved an accuracy of over 95% on the validation set, accurately extracting pesticide residue feature vectors from hyperspectral images.

[0044] The pesticide residue analysis subunit employs a support vector regression (SVR) model based on a radial basis function (RBF) kernel. Cross-validation is used to optimize model parameters, including the penalty parameter C and the width γ of the RBF kernel. During training, 20,000 sample data sets containing different pesticide types and residue concentrations were used to establish a nonlinear regression relationship between feature vectors and pesticide residues. The model achieved a mean square error (MSE) of less than 0.05, enabling accurate prediction of pesticide residues in agricultural products.

[0045] In addition to storing basic information, a pesticide-crop association table is also established, recording information such as the applicable crop range, recommended dosage, and safe interval for each pesticide. A data review mechanism is also in place, requiring newly entered pesticide information to be reviewed and confirmed by at least two professionals before it can be stored.

[0046] Establish a standards version management system to record the release date, revision history, and scope of application of standards in different countries and regions. Use crawler technology to monitor the websites of official agencies in various countries in real time. When standards are updated, the updated content is automatically downloaded and manually reviewed and verified to ensure the accuracy and completeness of the data.

[0047] Detailed records of crop biological classification information, including kingdom, phylum, class, order, family, genus, and species, are collected. At the same time, a correlation between crop growth environments and pesticide residue standards is established, such as differences in pesticide residue standards for the same crop under different climatic conditions, providing a more comprehensive basis for judging test results.

[0048] Utilizing multi-threaded parallel processing technology, upon receiving test results, multiple threads are simultaneously launched to retrieve corresponding standards from the standard limit database for different countries and regions, improving comparison efficiency. Standard priority rules are established to automatically prioritize standards based on information such as the agricultural product's sales destination and production location. A decision rule engine is built to configure flexible decision logic based on the standards requirements of different countries and regions. For single pesticide residue testing, test results are directly compared with the maximum residue limit (MRL) for the corresponding region. For complex determinations involving multiple pesticide residues, a comprehensive determination is made based on the synergistic and cumulative effects of the different pesticides, combined with appropriate risk assessment models. Results are categorized as qualified, unqualified, or pending retest. Test results close to the limit are marked as pending retest, prompting the user to conduct further testing and confirmation. In addition to graphical presentation, the system also provides a detailed test report generation function, which includes agricultural product information, test items, test results, standard limit values, analysis of any excesses, and the test rationale. Reports are available in various common file formats, including PDF and Excel, for convenient saving and printing.

[0049] The TCP / IP protocol is used as the basic communication protocol, and HTTP / HTTPS is used for data transmission at the application layer. For data with high real-time requirements (such as sensor data collection and test result output instructions), the WebSocket protocol is used for two-way communication to ensure fast data transmission and timely response. During the data transmission process, SSL / TLS encryption technology is used to encrypt the data to prevent data leakage and tampering, ensuring the security of data transmission.

[0050] The remote server utilizes a distributed architecture, deploying multiple server nodes to achieve load balancing and high availability. After establishing a connection with the server through the communication module, the detection terminal first undergoes identity authentication. Once authenticated, the detection data is packaged and uploaded to the server in the prescribed format. After receiving the data, the server performs data storage, backup, and preliminary analysis. Simultaneously, the server can push the latest model parameters, standard database updates, system upgrade patches, and other content to the detection terminal. Upon receiving this information, the detection terminal automatically performs the corresponding update operations to ensure the system remains up-to-date.

[0051] Furthermore, the detection system has added a terminal adaptation module and a cross-platform data service layer. The terminal adaptation module is responsible for identifying information such as the device type, operating system version, and screen resolution of mobile phones and computers, and formats the test result data according to the characteristics of different terminals. The cross-platform data service layer uses a RESTful API architecture as the data exchange interface between the detection system and mobile phone and computer applications, realizing unified data management and distribution.

[0052] Building on its existing functionality, the communication module optimizes the data transmission protocol to adapt to different terminals. For mobile phones, taking into account the complexity of the network environment and device performance limitations, a lightweight data compression algorithm (such as gzip) is used to compress test result data before transmission, reducing data traffic consumption and transmission time. For desktop computers, the speed and stability of data transmission are improved while ensuring data security. Furthermore, the WebSocket protocol is used to enable real-time push notifications for test results. When test results are generated or updated, the information is promptly pushed to mobile phones and desktop computers.

[0053] The mobile app features a simple and intuitive user interface based on the design specifications of Android and iOS systems. The main interface features a test results list, displaying key information such as the agricultural product name, test time, and determination result. Clicking an item in the list opens the detailed results page, showcasing basic test information, specific pesticide residue data, and comparison charts against standard limits using a combination of text and graphics. A responsive layout ensures optimal display on phones of varying screen sizes.

[0054] The desktop version is a web application developed using HTML5, CSS3, and JavaScript, supporting mainstream browsers. The main interface features a dashboard displaying test result statistics, such as the number of qualified and unqualified agricultural products and the percentage of test data from different regions. Clicking on a specific statistic opens a detailed data page, offering extensive filtering and sorting capabilities for easy analysis of test results. The interface is well-organized, taking advantage of the larger screen size of the computer to display more detailed information and data charts.

[0055] Strengthen security measures when viewing test results on mobile phones and computers. We use the OAuth2.0 protocol for user authentication and authorization management, ensuring that only legitimate users can access test result data. Data transmission is fully encrypted, including SSL / TLS encryption during communication and encryption when stored on terminal devices to prevent data theft or tampering. We also regularly scan and repair security vulnerabilities in mobile and computer applications to ensure secure and stable system operation.

[0056] A SPECIMIQ hyperspectral camera was selected, with a spectral resolution of 2.8 nm and a spatial resolution of 0.6 mm, meeting the system's detection requirements. The camera was mounted on a tripod with an adjustable height range of 0.5–2 m, and a pan / tilt system for 360° camera rotation. The computer was configured with an Intel Core i9-13900K processor, 64 GB of RAM, an NVIDIA GeForce RTX 4090 graphics card, and Windows 11 as the operating system. MySQL 8.0 was used as the database management software, and a local database server was established with a RAID 5 disk array to ensure data security and reliability. PyTorch 1.12 was selected as the deep learning framework, with CUDA 11.6 and cuDNN 8.3 installed to support GPU-accelerated computing.

[0057] During the data annotation phase, a team of 10 individuals with professional backgrounds in agricultural inspection was invited to form an annotation team. They used a multi-person annotation and cross-verification process to annotate the hyperspectral images to ensure accuracy. The 100,000 annotated images were divided into training, validation, and test sets in a ratio of 7:2:1. When training the convolutional neural network model, the batch size was set to 32, and evaluation was performed on the validation set after every five epochs. The learning rate and network parameters were adjusted based on the evaluation results. During support vector regression model training, a grid search method was used to search for the penalty parameter C in the range [0.1, 1000] and the width γ of the RBF kernel function in the range [0.001, 10]. The optimal parameter combination was determined through 10-fold cross-validation.

[0058] Place the agricultural product to be tested in the center of the stage. According to the type and size of the agricultural product, set the camera's scanning parameters (such as scanning speed, exposure time) and light source parameters (light intensity, spectral distribution) in the system operation interface. After clicking the Start Detection button, the system automatically controls the camera to scan, and the data acquisition module transmits the collected spectral image data to the data processing module in real time. The data processing module performs image preprocessing, feature extraction, and pesticide residue analysis in sequence, and transmits the analysis results to the judgment module. The judgment module compares and judges according to the preset standards, and finally the result output subunit generates a test report and displays it to the user. The judgment results are divided into three states: qualified, unqualified, and pending re-inspection. Test results close to the limit value are marked as pending re-inspection to remind the user to conduct further testing and confirmation. Users can view the test results on mobile phones and computers.

[0059] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. An artificial intelligence-based agricultural product pesticide residue detection system, characterized by: include: Data acquisition module, including a hyperspectral imaging unit and a data transmission unit; Acquire spectral image data of agricultural products through hyperspectral imaging technology and transmit the data to the data processing module; The data preprocessing module includes an image preprocessing subunit, a feature extraction subunit, and a pesticide residue analysis subunit, which performs preprocessing and feature extraction on the collected spectral image data; By building a deep learning model, the processed data is analyzed to determine the types and amounts of pesticide residues; The standard database module includes a pesticide variety database, a standard limit database, and a crop name database, which is used to store relevant information on pesticide varieties, standard limits, and crop names; The result output module includes a data comparison subunit that uses multi-threaded parallel processing technology and establishes standard priority rules, and a data result output subunit with the function of generating detailed test reports; it is used to compare the pesticide residue analysis results of the pesticide residue analysis subunit with the standards in the standard database module and output the test results; The communication module is used to realize data transmission between modules and connection with the remote server.

2. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that: The hyperspectral imaging unit in the data acquisition module uses a linear array CCD detector, a hyperspectral camera with a spectral range of 400-1000nm, a spectral resolution of 2.5nm and a spatial resolution of 0.1mm; the hyperspectral imaging unit is provided with an adjustable stage, the distance between the camera and the stage is maintained at 30-50cm, the hyperspectral camera is controlled by an electric slide to scan at a speed of 0.1-1m / s, and is provided with a halogen lamp and LED combination light source; the data transmission unit adopts a Gigabit Ethernet interface based on the TCP / IP protocol, is equipped with a data cache module and uses a compression algorithm; the data transmission unit compression ratio is between 1:10-1:

20.

3. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that: The data preprocessing module uses a three-dimensional block matching algorithm for denoising, a deep learning-based Retinex enhancement network to enhance images, and a polynomial correction method for geometric correction. It also introduces an attention mechanism module and a feature extraction subunit of a convolutional neural network model trained with a transfer learning strategy. The geometric correction of the image preprocessing subunit uses at least four ground control points, combined with camera intrinsic and extrinsic parameter information, to control the image geometric distortion error within 0.5 pixels.

4. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 3 is characterized in that: The subunits were extracted to construct an improved ResNet-50 convolutional neural network model. Based on the ImageNet pre-trained model, the model was fine-tuned on a dataset of 100,000 annotated hyperspectral images. The Adam optimizer was used, with the initial learning rate set to 0.001 and decayed by 0.1 every 10 epochs for 50 epochs.

5. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that: The pesticide residue analysis subunit uses a support vector regression model, trained using 20,000 sets of sample data, and optimized the penalty parameter C and RBF kernel function width γ through cross-validation. The model prediction mean square error is less than 0.

05.

6. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that: The crop name database records in detail the biological classification information of crops at the kingdom, phylum, class, order, family, genus and species; the standard limit database uses crawler technology to monitor the official websites of various countries or regions in real time, and automatically downloads when the standards are updated and stores them after manual review and verification.

7. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that the data The comparison subunit automatically determines the standard priority based on information such as the sales destination and production place of agricultural products; the test report generated by the result output subunit supports text report template generation, Echarts chart interactive display and local SQLite and cloud RDS data management.

8. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 7, characterized in that: The result output module is equipped with a judgment rule engine, which outputs judgment results in three states: qualified, unqualified and pending re-inspection.

9. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that: The communication module uses the TCP / IP protocol and WebSocket to transmit real-time data. The data is encrypted with SSL / TLS and supports interaction with distributed remote servers. The remote server adopts a distributed architecture and deploys multiple server nodes to achieve load balancing and high availability. After the detection terminal establishes a connection with the server through the communication module, it first performs identity authentication. After the authentication is passed, the detection data is packaged in the specified format and uploaded to the server. After receiving the data, the server performs data storage, backup, and preliminary analysis and processing; at the same time, the server can push the latest model parameters, standard database update information, and system upgrade patches to the detection terminal; After receiving the information pushed by the server, the detection terminal automatically performs the corresponding update operation to ensure that the system is always kept up to date.

10. The artificial intelligence-based agricultural product pesticide residue detection system according to claim 1, characterized in that: It also has a terminal adaptation module and a cross-platform data service layer. The terminal adaptation module is responsible for identifying the device type, operating system version, and screen resolution of mobile phones and computers, and formatting the test result data according to the characteristics of different terminals. The cross-platform data service layer adopts the RESTful API architecture as the data interaction interface between the detection system and mobile phone and computer applications, realizing unified management and distribution of data; the detection results can be viewed on mobile phones and computers.

Images