Farmland crop growth state intelligent monitoring method based on image recognition

Abstract

The invention discloses a farmland crop growth state intelligent monitoring method based on image recognition. The method comprises the following steps: S1, acquiring a farmland crop growth image by using ground acquisition equipment and preprocessing the farmland crop growth image; s2, crop image features of different spatial scales are extracted through a multi-scale convolution single page, and multi-level feature fusion is realized through residual connection; s3, adopting an improved firefly optimization algorithm to initialize a firefly group; s4, optimizing the structural parameters of the multi-scale residual network corresponding to each firefly individual; s5, identifying the standardized crop growth image data set; and S6, associating the crop growth state identification result with the corresponding crop geographical location information. According to the method, the improved firefly optimization algorithm, the multi-scale residual network and the image recognition technology are combined, and the farmland crop growth state intelligent monitoring based on image recognition is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural intelligent monitoring, and in particular to an intelligent monitoring method for the growth status of farmland crops based on image recognition. Background Art

[0002] With the rapid development of modern agriculture, monitoring the growth status of farmland crops has become a critical link in ensuring agricultural production efficiency and crop yields. Traditional methods for monitoring the growth status of farmland crops rely primarily on manual inspections, field sampling, and laboratory analysis. These methods are not only labor-intensive and time-consuming, but also significantly affected by human subjective factors, making it difficult to achieve large-scale, real-time, and efficient monitoring of farmland crop growth status. With the continuous development of emerging technologies such as remote sensing, the Internet of Things, and artificial intelligence, methods for monitoring the growth status of farmland crops based on image recognition and intelligent analysis have gradually become a hot topic in research and application. By processing and analyzing image data of farmland crops, not only can we achieve automated monitoring of crop growth status, health level, and spatial distribution, but we can also provide scientific decision-making support for agricultural production management.

[0003] Existing crop growth status monitoring methods based on image recognition primarily involve crop image acquisition, image preprocessing, feature extraction, state recognition and classification, and spatial distribution visualization. In recent years, with the advancement of computer vision and deep learning, various network architectures, such as convolutional neural networks and residual networks, have been widely used to identify and analyze crop growth status. Using deep network models to extract features from crop images effectively captures multi-dimensional features such as color, morphology, and texture at different growth stages, thereby improving the accuracy of crop state recognition. Furthermore, the introduction of methods such as multi-scale feature fusion and attention mechanisms has further enhanced the model's ability to discriminate crop growth status in complex farmland environments. Furthermore, some research has begun to focus on integrating crop status with geospatial information to visualize the spatial distribution of crop growth status in farmland, facilitating large-scale, dynamic agricultural monitoring.

[0004] However, existing technologies still have many shortcomings in practical applications. First, the image recognition models for crop growth status usually rely on manual experience or static settings in the selection of structural parameters, which makes it difficult to fully adapt to different crop varieties, different growth stages and changing farmland environments, thus affecting the generalization ability and practical application effect of the recognition model. Secondly, existing data enhancement methods are mostly performed in a single-channel or limited multi-channel manner, lacking efficient enhancement strategies for the diversity of crop images and complex backgrounds, resulting in insufficient robustness of the model in real scenarios. In addition, for the feature fusion and weighting of crop growth status, most methods only rely on simple splicing or averaging, lacking refined processing methods such as spatial attention mechanisms, making it difficult to highlight the role of key features in the final recognition. Furthermore, in terms of the association between crop growth status recognition results and geographic location, most existing methods use static mapping or manual labeling, failing to achieve efficient automatic association between recognition results and spatial information, affecting the accuracy and real-time performance of spatial distribution maps.

[0005] Furthermore, existing methods for generating spatial distribution maps and providing dynamic early warnings for crop growth status in farmland often rely on static regional divisions or manual statistical methods, lacking efficient spatial data processing and intelligent early warning mechanisms. If abnormal growth conditions occur within a farmland, early warning responses are delayed, making it difficult for agricultural authorities to provide accurate and dynamic management and intervention recommendations. Furthermore, existing optimization algorithms for model structure parameter optimization, such as genetic algorithms and particle swarm optimization, are prone to falling into local optimality in complex, high-dimensional parameter spaces. This results in limited search efficiency and accuracy, making them inadequate for the development of large-scale intelligent crop monitoring systems.

[0006] Therefore, how to provide an intelligent monitoring method for the growth status of farmland crops based on image recognition is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0007] One purpose of the present invention is to propose an intelligent monitoring method for the growth status of farmland crops based on image recognition. The present invention integrates the improved firefly optimization algorithm, multi-scale residual network and image recognition technology, and optimizes the network structure parameters by introducing strategies such as adaptive step size adjustment and optimal individual guidance, thereby significantly improving the model's ability to represent farmland crop images. The system adopts multi-scale feature fusion and spatial attention mechanism to fully explore the complex texture and spatial distribution characteristics of crop growth status, and realize high-precision status recognition and analysis. Combined with the automatic association of geographic spatial information and dynamic early warning module, it can generate a spatial distribution map of crop growth status in real time and make intelligent early warnings for abnormal changes. The overall solution has the capabilities of parameter adaptation, deep feature extraction and dynamic response, and can efficiently and intelligently complete crop growth status monitoring and management in complex farmland environments, thereby improving the scientific and intelligent level of agricultural production.

[0008] According to an embodiment of the present invention, a method for intelligently monitoring the growth status of farmland crops based on image recognition includes the following steps:

[0009] S1. Use ground acquisition equipment to collect farmland crop growth images and preprocess them to generate a standardized crop growth image dataset;

[0010] S2. Input the standardized crop growth image dataset into the multi-scale residual network, extract crop image features of different spatial scales through multi-scale convolution units, and achieve multi-level feature fusion through residual connections to generate crop growth status feature representation;

[0011] S3. Initialize the firefly population using the improved firefly optimization algorithm. Each firefly individual corresponds to a set of structural parameters of the multi-scale residual network.

[0012] S4. Using the recognition accuracy and loss value represented by the crop growth status characteristics as the fitness function, the multi-scale residual network structure parameters corresponding to each firefly individual are optimized, and the position of the firefly individual is iteratively updated to obtain the optimal structure parameters;

[0013] S5. Based on the optimal structural parameters, identify the standardized crop growth image dataset and output the crop growth status recognition result;

[0014] S6. Associate the crop growth status identification results with the corresponding crop geographical location information to generate a spatial distribution map of the farmland crop growth status, thereby realizing intelligent monitoring and dynamic early warning of the farmland crop growth status.

[0015] Optionally, the structural parameters of the multi-scale residual network specifically include the convolution kernel size of each convolution layer, the number of convolution kernels, the number of network layers, the connection method of each residual unit, the number of multi-scale branches, the pooling method and the step size.

[0016] Optionally, the crop growth status identification result specifically includes: the growth stage, health status, and growth index of each crop.

[0017] Optionally, the S2 specifically includes:

[0018] S21. Input the standardized crop growth image dataset into the input layer of the multi-scale residual network to obtain the initial feature representation of the image;

[0019] S22, inputting the initial image feature representation into multi-scale convolution units with different convolution kernel sizes, extracting crop image features at different spatial scales, and obtaining a multi-scale crop feature representation;

[0020] S23, inputting the multi-scale crop feature representations into the residual unit respectively, fusing the multi-scale features through the residual connection, and obtaining a preliminary fused feature representation;

[0021] S24, performing batch normalization and activation function processing on the preliminary fused feature representation to obtain a normalized fused feature representation;

[0022] S25, inputting the normalized fusion feature representation into the global pooling layer to obtain a global crop growth state feature representation;

[0023] S26. Output the global crop growth state feature representation as the crop growth state feature representation.

[0024] Optionally, the S3 specifically includes:

[0025] S31. Setting the search space and optimization target for the multi-scale residual network structural parameters, generating a firefly swarm with an initial population size, and introducing an improved firefly optimization algorithm. The improved firefly optimization algorithm includes an adaptive step-size adjustment mechanism, which dynamically adjusts the firefly displacement step size to balance the global search and local search capabilities. At the same time, an optimal individual guidance strategy is introduced to enhance the population's convergence speed to a high-quality solution, thereby improving the search efficiency and result accuracy of the optimal structural parameters.

[0026] S32, randomly assigning a set of multi-scale residual network structure parameters to each firefly individual to obtain an initial firefly individual parameter set;

[0027] S33, evaluating the fitness of each firefly individual according to the multi-scale residual network structure parameters to obtain the fitness value of the firefly individual;

[0028] S34, according to the individual firefly fitness value and distance, using the improved individual firefly movement strategy to update the individual firefly structure parameters, and obtain an updated individual firefly parameter set;

[0029] S35: Determine whether the number of iterations reaches 100, or the fitness of the best individual in the firefly population is improved by less than 0.001. If not, use the updated firefly individual parameter set as input for a new round of optimization, and return to step S33;

[0030] S36. If the termination condition is met, the structural parameters of the optimal individual in the current firefly population are output as the initialization structural parameters of the multi-scale residual network.

[0031] Optionally, the S4 specifically includes:

[0032] S41. Generate an initial firefly individual parameter set based on the initialized structural parameters of the multi-scale residual network;

[0033] S42. Using the multi-scale residual network structure parameters of each firefly individual, respectively train and identify the crop growth state feature representation, and obtain the recognition accuracy and loss value corresponding to each firefly individual;

[0034] S43, using the recognition accuracy and loss value as the fitness function, calculating the fitness value of each firefly individual to obtain a fitness value set;

[0035] S44, updating the individual structural parameters of the fireflies using an improved firefly optimization algorithm according to the fitness value set to obtain an updated individual parameter set of the fireflies;

[0036] S45, determining whether the number of iterations has reached 100; if not, using the updated firefly individual parameter set as input for a new round of optimization, and returning to step S42;

[0037] S46. If the preset termination condition is reached, the structural parameters of the individual with the highest fitness value in the current firefly population are output as the optimal structural parameters.

[0038] Optionally, the S5 specifically includes:

[0039] S51. Apply the optimal structural parameters to the multi-scale residual network, and adaptively adjust the network input layer structure based on the specific variety information of the current farmland crops;

[0040] S52, performing multi-channel data enhancement processing on the collected crop image data to obtain enhanced crop image data;

[0041] S53, inputting the enhanced crop image data into a multi-scale residual network that has been adaptively adjusted to extract a multi-scale feature representation of crop growth status;

[0042] S54, performing feature fusion on the extracted crop growth status feature representation, combining the spatial attention mechanism to perform feature weighting, and obtaining a weighted comprehensive feature representation;

[0043] S55. Utilize the weighted comprehensive feature representation and adopt a multi-category discrimination strategy to classify and identify the crop growth status and generate a growth status identification result.

[0044] Optionally, the S6 specifically includes:

[0045] S61, processing the geographical location information of each crop plot in the farmland area to generate geographical location data;

[0046] S62, associating the growth status recognition result with the geographic location data to form growth status spatial annotation data;

[0047] S63, performing spatial coordinate conversion on the growth status spatial annotation data to achieve geographic mapping of the crop growth status;

[0048] S64. Based on the spatially labeled data of the growth status after geographic mapping, construct a draft of the spatial distribution map of the growth status of farmland crops;

[0049] S65. Perform spatial analysis and visualization on the draft spatial distribution map to generate a high-precision spatial distribution map of crop growth status in farmland.

[0050] S66. Set a dynamic warning threshold based on the spatial distribution map, identify areas with abnormal growth status, and output dynamic warning information on the growth status of farmland crops.

[0051] The beneficial effects of the present invention are:

[0052] The present invention significantly improves the efficiency and accuracy of intelligent monitoring of crop growth status in farmland by organically combining the improved firefly optimization algorithm, multi-scale residual network and image recognition technology. First, the adaptive optimization of structural parameters greatly enhances the generalization ability of the model in different crop varieties, different growth stages and variable farmland environments, overcoming the problem of insufficient adaptability caused by the traditional monitoring method relying on manual experience and parameter solidification. Secondly, the introduction of multi-scale residual network and spatial attention mechanism enables the model to fully extract and fuse the multi-level and multi-dimensional features of crop images, effectively improving the discrimination ability and robustness of crop growth status under complex backgrounds. In addition, the combination of multi-channel image enhancement and automatic association technology of geographic spatial information not only expands the model's perception dimension of crop information, but also realizes real-time visualization of crop growth status recognition results and spatial distribution, improving the data processing efficiency and spatial expression ability of large-scale farmland monitoring.

[0053] More importantly, the present invention realizes the automatic detection and real-time alarm of abnormal crop growth status through the generation of dynamic spatial distribution maps and intelligent early warning mechanisms, which greatly improves the timeliness and scientificity of agricultural production management. When the abnormal growth status of crops is discovered, the system can immediately associate the abnormal information with the specific geographical location, assisting managers to quickly locate risk areas and take targeted measures. This not only improves the refinement and intelligence level of farmland management, but also provides strong support for improving the quality and efficiency of agricultural production. Overall, the present invention overcomes the limitations of the existing technology in parameter optimization, feature extraction, spatial information processing and dynamic early warning, and realizes the efficient, real-time and automated intelligent monitoring of the growth status of farmland crops, which has important practical value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 This is a flow chart of an intelligent monitoring method for crop growth status in farmland based on image recognition proposed by the present invention;

[0056] Figure 2 This is a schematic diagram of an intelligent monitoring method for crop growth status in farmland based on image recognition proposed by the present invention;

[0057] Figure 3 This is a data flow diagram of an intelligent monitoring method for farmland crop growth status based on image recognition proposed by the present invention. DETAILED DESCRIPTION

[0058] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0059] refer to Figure 1-3 , an intelligent monitoring method for crop growth status in farmland based on image recognition, comprising the following steps:

[0060] S1. Use ground acquisition equipment to collect farmland crop growth images and preprocess them to generate a standardized crop growth image dataset;

[0061] S2. Input the standardized crop growth image dataset into the multi-scale residual network, extract crop image features of different spatial scales through multi-scale convolution units, and achieve multi-level feature fusion through residual connections to generate crop growth status feature representation;

[0062] S3. Initialize the firefly population using the improved firefly optimization algorithm. Each firefly individual corresponds to a set of structural parameters of the multi-scale residual network.

[0063] S4. Using the recognition accuracy and loss value represented by the crop growth status characteristics as the fitness function, the multi-scale residual network structure parameters corresponding to each firefly individual are optimized, and the position of the firefly individual is iteratively updated to obtain the optimal structure parameters;

[0064] S5. Based on the optimal structural parameters, identify the standardized crop growth image dataset and output the crop growth status recognition result;

[0065] S6. Associate the crop growth status identification results with the corresponding crop geographical location information to generate a spatial distribution map of the farmland crop growth status, thereby realizing intelligent monitoring and dynamic early warning of the farmland crop growth status.

[0066] This invention combines a multiscale residual network with an improved firefly optimization algorithm to achieve efficient and accurate identification of crop growth status in farmland. By utilizing multi-level feature fusion and adaptive optimization of structural parameters, the model improves its ability to discriminate crop growth status in complex field environments. Combining spatial distribution with dynamic early warning effectively enhances the real-time and scientific nature of intelligent farmland monitoring.

[0067] In this embodiment, the structural parameters of the multi-scale residual network specifically include the convolution kernel size of each convolution layer, the number of convolution kernels, the number of network layers, the connection method of each residual unit, the number of multi-scale branches, the pooling method and the step size.

[0068] This method achieves adaptive adjustment of the multiscale residual network structure by optimizing structural parameters such as convolution kernel size, number of convolution kernels, number of network layers, residual unit connection method, number of multiscale branches, and pooling method. This method can flexibly match the optimal parameter combination for different crop image features, improving the diversity and discriminative power of feature extraction, thereby significantly improving the accuracy and robustness of crop growth status recognition.

[0069] In this embodiment, the crop growth status identification result specifically includes: the growth stage, health status, and growth index of each crop.

[0070] This method achieves comprehensive and precise identification of crop growth status by outputting the growth stage, health status, and growth indicators of each crop. This method can provide a multi-dimensional, quantifiable decision-making basis for precision agriculture management, improving the targeted and scientific nature of pest and disease early warning and field management, and effectively contributing to high-yield and high-quality crops and intelligent farmland management.

[0071] In this embodiment, S2 specifically includes:

[0072] S21. Input the standardized crop growth image dataset into the input layer of the multi-scale residual network to obtain the initial feature representation of the image;

[0073] S22, inputting the initial image feature representation into multi-scale convolution units with different convolution kernel sizes, extracting crop image features at different spatial scales, and obtaining a multi-scale crop feature representation;

[0074] S23, inputting the multi-scale crop feature representations into the residual unit respectively, fusing the multi-scale features through the residual connection, and obtaining a preliminary fused feature representation;

[0075] S24, performing batch normalization and activation function processing on the preliminary fused feature representation to obtain a normalized fused feature representation;

[0076] S25, inputting the normalized fusion feature representation into the global pooling layer to obtain a global crop growth state feature representation;

[0077] S26. Output the global crop growth state feature representation as the crop growth state feature representation.

[0078] This method uses multi-scale convolutional units to extract crop image features at different spatial scales and combines them with residual connections to achieve effective fusion of these multi-scale features. After batch normalization and global pooling, a highly expressive global representation of crop growth status is obtained. This method fully exploits multi-level image information, improving the comprehensiveness of feature extraction and the accuracy of discrimination, providing a solid foundation for subsequent crop growth status identification.

[0079] In this embodiment, S3 specifically includes:

[0080] S31. Setting the search space and optimization target for the multi-scale residual network structural parameters, generating a firefly swarm with an initial population size, and introducing an improved firefly optimization algorithm. The improved firefly optimization algorithm includes an adaptive step-size adjustment mechanism, which dynamically adjusts the firefly displacement step size to balance the global search and local search capabilities. At the same time, an optimal individual guidance strategy is introduced to enhance the population's convergence speed to a high-quality solution, thereby improving the search efficiency and result accuracy of the optimal structural parameters.

[0081] S32, randomly assigning a set of multi-scale residual network structure parameters to each firefly individual to obtain an initial firefly individual parameter set;

[0082] S33, evaluating the fitness of each firefly individual according to the multi-scale residual network structure parameters to obtain the fitness value of the firefly individual;

[0083] S34, according to the individual firefly fitness value and distance, using the improved individual firefly movement strategy to update the individual firefly structure parameters, and obtain an updated individual firefly parameter set;

[0084] S35: Determine whether the number of iterations reaches 100, or the fitness of the best individual in the firefly population is improved by less than 0.001. If not, use the updated firefly individual parameter set as input for a new round of optimization, and return to step S33;

[0085] S36. If the termination condition is met, the structural parameters of the optimal individual in the current firefly population are output as the initialization structural parameters of the multi-scale residual network.

[0086] This paper introduces an adaptive step-size adjustment mechanism and an optimal individual guidance strategy to enhance the global exploration and local convergence capabilities of the improved Firefly Optimization Algorithm during the multi-scale residual network structural parameter search process. This method dynamically balances the search range and convergence speed, effectively accelerating the search for optimal structural parameters and improving the accuracy and stability of network structural parameter optimization results, providing a solid foundation for subsequent crop growth status identification.

[0087] In this embodiment, the S4 specifically includes:

[0088] S41. Generate an initial firefly individual parameter set based on the initialized structural parameters of the multi-scale residual network;

[0089] S42. Using the multi-scale residual network structure parameters of each firefly individual, respectively train and identify the crop growth state feature representation, and obtain the recognition accuracy and loss value corresponding to each firefly individual;

[0090] S43, using the recognition accuracy and loss value as the fitness function, calculating the fitness value of each firefly individual to obtain a fitness value set;

[0091] S44, updating the individual structural parameters of the fireflies using an improved firefly optimization algorithm according to the fitness value set to obtain an updated individual parameter set of the fireflies;

[0092] S45, determining whether the number of iterations has reached 100; if not, using the updated firefly individual parameter set as input for a new round of optimization, and returning to step S42;

[0093] S46. If the preset termination condition is reached, the structural parameters of the individual with the highest fitness value in the current firefly population are output as the optimal structural parameters.

[0094] This method utilizes a dual fitness function—recognition accuracy and loss—combined with an improved Firefly Optimization algorithm to dynamically iterate network structure parameters, effectively improving the globality and accuracy of parameter optimization. Through rounds of training and adaptive optimization, the selected network structure achieves higher accuracy and lower loss in crop growth status recognition tasks, achieving continuous improvement in model performance and stable output.

[0095] In this embodiment, the S5 specifically includes:

[0096] S51. Apply the optimal structural parameters to the multi-scale residual network, and adaptively adjust the network input layer structure based on the specific variety information of the current farmland crops;

[0097] S52, performing multi-channel data enhancement processing on the collected crop image data to obtain enhanced crop image data;

[0098] S53, inputting the enhanced crop image data into a multi-scale residual network that has been adaptively adjusted to extract a multi-scale feature representation of crop growth status;

[0099] S54, performing feature fusion on the extracted crop growth status feature representation, combining the spatial attention mechanism to perform feature weighting, and obtaining a weighted comprehensive feature representation;

[0100] S55. Utilize the weighted comprehensive feature representation and adopt a multi-category discrimination strategy to classify and identify the crop growth status and generate a growth status identification result.

[0101] This method adaptively adjusts the network structure based on crop variety information and enhances sample diversity and model generalization through multi-channel data augmentation. It also incorporates a spatial attention mechanism to weight multi-scale features, effectively highlighting key growth characteristics. It also employs a multi-class discrimination strategy to improve the accuracy and stability of crop growth status recognition, enabling efficient and intelligent recognition in complex farmland environments.

[0102] In this embodiment, S6 specifically includes:

[0103] S61, processing the geographical location information of each crop plot in the farmland area to generate geographical location data;

[0104] S62, associating the growth status recognition result with the geographic location data to form growth status spatial annotation data;

[0105] S63, performing spatial coordinate conversion on the growth status spatial annotation data to achieve geographic mapping of the crop growth status;

[0106] S64. Based on the spatially labeled data of the growth status after geographic mapping, construct a draft of the spatial distribution map of the growth status of farmland crops;

[0107] S65. Perform spatial analysis and visualization on the draft spatial distribution map to generate a high-precision spatial distribution map of crop growth status in farmland.

[0108] S66. Set a dynamic warning threshold based on the spatial distribution map, identify areas with abnormal growth status, and output dynamic warning information on the growth status of farmland crops.

[0109] This method links crop growth status with geographic location information, enabling spatial visualization and precise mapping of growth status. Using spatial analysis and dynamic early warning threshold setting, it effectively identifies and locates areas of abnormal growth, enhancing intelligent farmland management and early warning capabilities. This method supports the generation of high-precision spatial distribution maps, providing a scientific basis for precision agriculture decision-making and abnormal intervention.

[0110] Example 1:

[0111] To verify the feasibility of the present invention in practice, it was applied to the intelligent monitoring of wheat growth status in a modern agricultural industrial park. This area is typical of the Jianghuai Plain farmland, with a planting area of ​​approximately 1,200 mu (approximately 1,000 acres) covering different plots, crop varieties, and diversified management models. Previously, crop growth status monitoring in this area relied primarily on regular field inspections and manual sampling and analysis by agricultural technicians. This approach presented numerous issues, including long data collection cycles, limited coverage, and large subjective errors. Furthermore, it was difficult to promptly detect and locate local growth anomalies, hindering refined management and yield improvement.

[0112] In this example, we deployed a high-resolution drone remote sensing image acquisition system to regularly capture images of wheat fields within the park. The acquisition frequency was once every seven days, and over 7,200 images were collected throughout the entire growing season. The captured images first underwent multi-channel enhancement processing in the data preprocessing module, effectively mitigating image quality fluctuations caused by environmental interference such as lighting, shadows, and weeds. The enhanced images were input into a multi-scale residual network. By fusing spatial features at different scales and introducing a spatial attention mechanism, the model was able to automatically distinguish between different target objects, such as wheat plants, weeds, and soil, and identify the growth stage and health status of the wheat.

[0113] An improved Firefly optimization algorithm was used to optimize model parameters. Building on the standard algorithm, this algorithm incorporates an adaptive step-size adjustment mechanism and an optimal individual guidance strategy. This allows the model to dynamically adjust parameters for improved recognition performance in diverse farmland environments and complex image features. Compared to traditional manual parameter tuning or static optimization algorithms, the model's convergence speed increased by approximately 25%, and the final recognition accuracy reached 95.3%, a 5 percentage point improvement over conventional convolutional neural networks. Furthermore, integrated with a geographic information system (GIS) module, the system automatically correlates the recognition results of each image with its acquisition coordinates, creating a real-time spatial distribution map of crop growth status. Monitoring results show that from April to October 2023, the system detected and located 42 wheat lodging incidents, 36 disease incidents, and 28 nitrogen deficiency incidents. This spatial distribution map allows managers to immediately identify the specific plots and areas of abnormal distribution, effectively guiding management measures such as targeted fertilization and pest control.

[0114] In actual application, the intelligent early warning module of the present invention fully demonstrates the advantages of dynamic response. On May 16, 2023, during the wheat heading period, the system detected abnormal plant growth in two consecutive monitoring periods in Block D. After spatial distribution map analysis, it was found that about 3.8 acres of land in the block showed nitrogen deficiency signals with yellow leaves and short plant height. The system automatically generates early warning information and pushes it to agricultural technicians through mobile terminals. After receiving the early warning, the management personnel quickly went to the site for verification and applied nitrogen fertilizer in time according to the monitoring recommendations. Subsequent tracking showed that the growth of wheat in the block quickly returned to normal, and the yield loss was controlled at 1.7%, which was a significant improvement compared to the average loss of more than 6% for similar problems in previous years.

[0115] Furthermore, the system identified suspected early symptoms of wheat stripe rust in Block F, covering approximately 2.1 mu (approximately 2.1 acres). This warning information was simultaneously pushed to the management platform, enabling technicians to promptly implement preventive measures and effectively curb the spread of the disease. Statistics show that the disease incidence in the controlled block was ultimately limited to 0.8%, while the incidence in the control area, which did not utilize the system, was as high as 5.2% at the same stage. During the entire monitoring period, the system issued 106 warnings regarding abnormal growth conditions, reducing the average management response time from 48 hours to less than 6 hours, significantly improving farmland management efficiency.

[0116] Table 1 Comparison of optimization effects of intelligent monitoring of crop growth status based on image recognition

[0117] Performance indicators System of the present invention Traditional methods Recognition accuracy (%) 95.3 90.2 Recognition time (single block / minute) 0.2 6.8 Spatial distribution map accuracy (%) 98.7 81.5 Abnormal warning response time (h) 5.6 48 Annual output increase (%) 4.6 0 Improvement in economic benefits per mu (yuan) 368 0 Reduction in labor input (%) 72 0 Disease incidence (%) 0.8 5.2 Yield loss due to nitrogen deficiency (%) 1.7 6.1 Inspection frequency (times / month) 1 4

[0118] Table 1 shows that after the application of the present invention, the overall wheat yield in the region increased by 4.6% compared with the previous year, and the average economic benefit per mu increased by about 368 yuan. At the same time, the frequency of management personnel inspections was reduced from 4 times a month to more than 1 time a month, and manual input was reduced by 72%. The average time required for system identification and spatial distribution map generation was shortened from 220 minutes of manual marking to less than 15 minutes, and the accuracy of the spatial distribution map reached 98.7%. The above data fully demonstrate that the present invention significantly improves the automation, intelligence and high-precision level of farmland crop growth status monitoring, effectively solves the shortcomings of traditional methods in timeliness, accuracy and spatial positioning, and provides a solid technical support and data foundation for modern agricultural production.

[0119] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An intelligent monitoring method for crop growth status in farmland based on image recognition, characterized in that: The steps include: S1. Use ground acquisition equipment to collect farmland crop growth images and preprocess them to generate a standardized crop growth image dataset; S2. Input the standardized crop growth image dataset into the multi-scale residual network, extract crop image features of different spatial scales through multi-scale convolution units, and achieve multi-level feature fusion through residual connections to generate crop growth status feature representation; S3. Initialize the firefly population using the improved firefly optimization algorithm. Each firefly individual corresponds to a set of structural parameters of the multi-scale residual network. S4. Using the recognition accuracy and loss value represented by the crop growth status characteristics as the fitness function, the multi-scale residual network structure parameters corresponding to each firefly individual are optimized, and the position of the firefly individual is iteratively updated to obtain the optimal structure parameters; S5. Based on the optimal structural parameters, identify the standardized crop growth image dataset and output the crop growth status recognition result; S6. Associate the crop growth status identification results with the corresponding crop geographical location information to generate a spatial distribution map of the farmland crop growth status, thereby realizing intelligent monitoring and dynamic early warning of the farmland crop growth status.

2. The method for intelligent monitoring of crop growth status based on image recognition according to claim 1, characterized in that: The structural parameters of the multi-scale residual network specifically include the convolution kernel size of each convolution layer, the number of convolution kernels, the number of network layers, the connection method of each residual unit, the number of multi-scale branches, the pooling method and the step size.

3. The method for intelligent monitoring of crop growth status based on image recognition according to claim 1, characterized in that: The crop growth status identification result specifically includes: the growth stage, health status, and growth index of each crop.

4. The method for intelligent monitoring of crop growth status based on image recognition according to claim 1, characterized in that: The S2 specifically includes: S21. Input the standardized crop growth image dataset into the input layer of the multi-scale residual network to obtain the initial feature representation of the image; S22, inputting the initial image feature representation into multi-scale convolution units with different convolution kernel sizes, extracting crop image features at different spatial scales, and obtaining a multi-scale crop feature representation; S23, inputting the multi-scale crop feature representations into the residual unit respectively, fusing the multi-scale features through the residual connection, and obtaining a preliminary fused feature representation; S24, performing batch normalization and activation function processing on the preliminary fused feature representation to obtain a normalized fused feature representation; S25, inputting the normalized fusion feature representation into the global pooling layer to obtain a global crop growth state feature representation; S26. Output the global crop growth state feature representation as the crop growth state feature representation.

5. The method for intelligent monitoring of crop growth status based on image recognition according to claim 1, characterized in that: The S3 specifically includes: S31. Setting the search space and optimization target for the multi-scale residual network structural parameters, generating a firefly swarm with an initial population size, and introducing an improved firefly optimization algorithm. The improved firefly optimization algorithm includes an adaptive step-size adjustment mechanism, which dynamically adjusts the firefly displacement step size to balance the global search and local search capabilities. At the same time, an optimal individual guidance strategy is introduced to enhance the population's convergence speed to a high-quality solution, thereby improving the search efficiency and result accuracy of the optimal structural parameters. S32, randomly assigning a set of multi-scale residual network structure parameters to each firefly individual to obtain an initial firefly individual parameter set; S33, evaluating the fitness of each firefly individual according to the multi-scale residual network structure parameters to obtain the fitness value of the firefly individual; S34, according to the individual firefly fitness value and distance, using the improved individual firefly movement strategy to update the individual firefly structure parameters, and obtain an updated individual firefly parameter set; S35: Determine whether the number of iterations reaches 100, or the fitness of the best individual in the firefly population is improved by less than 0.

001. If not, use the updated firefly individual parameter set as input for a new round of optimization, and return to step S33; S36. If the termination condition is met, the structural parameters of the optimal individual in the current firefly population are output as the initialization structural parameters of the multi-scale residual network.

6. The method for intelligently monitoring the growth status of farmland crops based on image recognition according to claim 1, characterized in that: The S4 specifically includes: S41. Generate an initial firefly individual parameter set based on the initialized structural parameters of the multi-scale residual network; S42. Using the multi-scale residual network structure parameters of each firefly individual, respectively train and identify the crop growth state feature representation, and obtain the recognition accuracy and loss value corresponding to each firefly individual; S43, using the recognition accuracy and loss value as the fitness function, calculating the fitness value of each firefly individual to obtain a fitness value set; S44, updating the individual structural parameters of the fireflies using an improved firefly optimization algorithm according to the fitness value set to obtain an updated individual parameter set of the fireflies; S45, determining whether the number of iterations has reached 100; if not, using the updated firefly individual parameter set as input for a new round of optimization, and returning to step S42; S46. If the preset termination condition is reached, the structural parameters of the individual with the highest fitness value in the current firefly population are output as the optimal structural parameters.

7. The method for intelligently monitoring crop growth status based on image recognition according to claim 1, characterized in that: The S5 specifically includes: S51. Apply the optimal structural parameters to the multi-scale residual network, and adaptively adjust the network input layer structure based on the specific variety information of the current farmland crops; S52, performing multi-channel data enhancement processing on the collected crop image data to obtain enhanced crop image data; S53, inputting the enhanced crop image data into a multi-scale residual network that has been adaptively adjusted to extract a multi-scale feature representation of crop growth status; S54, performing feature fusion on the extracted crop growth status feature representation, combining the spatial attention mechanism to perform feature weighting, and obtaining a weighted comprehensive feature representation; S55. Utilize the weighted comprehensive feature representation and adopt a multi-category discrimination strategy to classify and identify the crop growth status and generate a growth status identification result.

8. The method for intelligent monitoring of crop growth status based on image recognition according to claim 1, characterized in that: The S6 specifically includes: S61, processing the geographical location information of each crop plot in the farmland area to generate geographical location data; S62, associating the growth status recognition result with the geographic location data to form growth status spatial annotation data; S63, performing spatial coordinate conversion on the growth status spatial annotation data to achieve geographic mapping of the crop growth status; S64. Based on the spatially labeled data of the growth status after geographic mapping, construct a draft of the spatial distribution map of the growth status of farmland crops; S65. Perform spatial analysis and visualization on the draft spatial distribution map to generate a high-precision spatial distribution map of crop growth status in farmland. S66. Set a dynamic warning threshold based on the spatial distribution map, identify areas with abnormal growth status, and output dynamic warning information on the growth status of farmland crops.

Images