Insect pest image recognition data processing method and platform

Abstract

The invention discloses an insect pest image recognition data processing method and platform, and aims to solve the problem that a traditional insect pest recognition method often depends on a single environmental condition or seasonal data and is easy to cause performance reduction of a model in different environments, while the problem can be effectively solved through data acquisition and enhanced multi-source data introduction. And the generalization ability of the system is improved. And secondly, through occlusion recovery, region speculation and seasonal change adaptive optimization, the system can process occlusion and seasonal change in the image in real time, and the recognition precision is further improved. According to multi-scale feature fusion and data coupling, different types of image information are comprehensively integrated, so that the accuracy of an identification result is effectively improved. And finally, through an instant feedback mechanism, an operator can adjust and optimize a recognition result more quickly, and the adaptability and efficiency of the system are further improved.

Description

Technical Field

[0001] The present application relates to the field of agricultural technology, and in particular to a method and platform for processing pest image recognition data. Background Art

[0002] As the demand for efficient and accurate detection in agricultural production continues to grow, image recognition technology has become a key tool in pest control. This technology uses cameras to capture pest images and then utilizes deep learning algorithms to classify and identify the pests in these images, enabling automated pest monitoring and management. This approach allows farmers to efficiently and accurately identify pest types and distribution, enabling them to implement effective prevention and control measures in a timely manner, significantly improving pest control efficiency and reducing manual inspection costs.

[0003] Although pest image recognition technology has made significant progress, there are still some urgent problems to be solved in practical applications. First, background complexity seriously affects recognition accuracy. Pest images usually have color, texture and other features that are highly similar to the background, which increases the risk of misidentification. Second, seasonal changes pose a challenge to image recognition capabilities. As the seasons change, the morphology, color and distribution of plants and pests will change significantly, and existing models often lack sufficient adaptability. Finally, object occlusion and pest overlap are common problems in image recognition, especially when the pest is partially obscured by plant leaves or branches. Existing recognition technologies perform poorly when dealing with occluded areas. Summary of the Invention

[0004] The main purpose of this application is to provide a pest image recognition data processing method and platform to solve the problems raised by the above background technology.

[0005] To achieve the above objectives, this application provides the following technical solutions:

[0006] A method for processing pest image recognition data, comprising the following specific steps:

[0007] S1. Data collection and enhancement: collecting pest image data from different environments, seasons, and lighting conditions, and preprocessing and reorganizing the collected multi-source data to generate the first, second, and third data sets;

[0008] S2, occlusion recovery and area estimation, coupling the first data group, the second data group, and the third data group to generate an occlusion anomaly coefficient ZDY, performing data analysis on the occlusion anomaly coefficient ZDY, and determining whether an occlusion anomaly problem occurs based on the analysis results, and generating an occlusion intervention coefficient ZDJ based on the analysis results;

[0009] S3, seasonal variation adaptability optimization, coupling the first data group, the second data group, and the third data group to generate a seasonal adaptation coefficient JJS;

[0010] S4. Multi-scale feature fusion: coupling the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS of the first, second, and third data groups to generate an image pest recognition coefficient RXS. Data analysis is performed on the image pest recognition coefficient RXS, and based on the analysis results, it is determined whether there are pests in the image area.

[0011] S5. Feedback: Feedback various parameters to the visualization terminal.

[0012] Preferably, in step S1, the specific steps of data collection and enhancement are as follows:

[0013] S1.1. Use image analysis equipment to collect image data, including obstruction area ratio, obstruction depth, number of obstructed objects, obstruction area density, obstruction area contrast, seasonal light intensity, light variation amplitude, seasonal background variation, pest growth cycle, ambient temperature index, pest area, pest contrast, edge clarity, pest boundary density, and pest obstruction degree;

[0014] S1.2, pre-process the collected parameters through the image processing system and make them dimensionless;

[0015] S1.3. Rearrange the parameters after preprocessing to generate a first data set, a second data set, and a third data set;

[0016] The first data set includes occlusion area ratio A, occlusion depth B, number of occluding objects C, occlusion area density D, and occlusion area contrast E;

[0017] The second data set includes seasonal light intensity F, light variation G, seasonal background variation H, pest growth cycle I, and ambient temperature index G;

[0018] The third data set includes the pest area K, pest contrast L, edge definition M, pest boundary density N, and pest occlusion O.

[0019] Preferably, in step S2, the specific steps of occlusion recovery and area estimation are as follows:

[0020] S2.1. Extract parameters from the first, second, and third data sets, including occlusion area ratio A, occlusion depth B, number of occluding objects C, occlusion area density D, seasonal light intensity F, seasonal background variation H, pest occlusion O, and edge definition M. Input these extracted parameters into a pre-trained deep learning framework, perform feature fusion on them through a multi-layer neural network, and calculate the occlusion anomaly coefficient ZDY.

[0021] S2.2. Perform data analysis on the occlusion anomaly coefficient ZDY to generate analysis results. Based on the analysis results, determine whether to include it in subsequent steps. The specific analysis results are as follows:

[0022] When ZDY≤0.5, it means that the occlusion anomaly coefficient ZDY does not need to be added to the subsequent calculations;

[0023] When ZDY>0.5, it means that the occlusion anomaly coefficient ZDY needs to be added to the subsequent calculations;

[0024] S2.3. Parameters are extracted from the first data group, the second data group, and the third data group, including the occlusion area ratio A, the occlusion depth B, the occlusion area density D, the seasonal light intensity F, the seasonal background change H, the pest occlusion degree O, and the pest boundary density N. The extracted multiple parameters are input into the pre-trained deep learning framework, and the features are fused through a multi-layer neural network to calculate and generate the occlusion intervention coefficient ZDJ.

[0025] Preferably, in step S2.2, the occlusion anomaly coefficient ZDY is calculated by the following formula:

[0026]

[0027] Where: A is the occlusion area ratio, B is the occlusion depth, C is the number of occluding objects, D is the occlusion area density, F is the seasonal light intensity, H is the seasonal background change, O is the pest occlusion degree, M is the edge clarity, a1 and a2 are weight values, and the values of a1 and a2 are adjusted by the user.

[0028] Preferably, in step S2.3, the occlusion intervention coefficient ZDJ is calculated by the following formula:

[0029]

[0030] Where: A is the shading area ratio, B is the shading depth, D is the shading area density, F is the seasonal light intensity, H is the seasonal background change, O is the pest shading degree, N is the pest boundary density, b1 and b2 are weight values, and the values of b1 and b2 are adjusted by the user.

[0031] Preferably, in step S3, parameters are extracted from the first data group, the second data group, and the third data group, including seasonal light intensity F, seasonal background change H, ambient temperature index G, pest area K, shading area ratio A, pest shading degree O, and shading area density D. The extracted multiple parameters are input into a pre-trained deep learning framework, and feature fusion is performed through a multi-layer neural network to calculate and generate the seasonal adaptation coefficient JJS. The specific calculation method is as follows:

[0032]

[0033] Where: F is the seasonal light intensity, H is the seasonal background change, G is the ambient temperature index, K is the area of the pest area, A is the shading area ratio, O is the pest shading degree, D is the shading area density, c1 and c2 are weight values, and the values of b1 and b2 are adjusted by the user.

[0034] Preferably, in step S4, the specific steps of multi-scale feature fusion are as follows:

[0035] S4.1. Extract parameters from the first, second, and third data sets, including the obscured area ratio A, obscured area contrast E, seasonal light intensity F, pest area K, and pest growth period I. Input these extracted parameters, along with the obscured intervention coefficient ZDJ and the seasonal adaptation coefficient JJS, into a pre-trained deep learning framework. Feature fusion is performed using a multi-layer neural network to calculate the image pest recognition coefficient RXS.

[0036] S4.2. Perform data analysis on the image pest identification coefficient RXS to generate an analysis result. Based on the analysis result, determine whether there is a pest problem in the current image area.

[0037] Preferably, in step S4.1, the specific calculation formulas for the image pest recognition coefficient RXS are as follows:

[0038]

[0039] Where A is the shaded area ratio, E is the shaded area contrast, F is the seasonal light intensity, K is the area of the pest area, I is the pest growth cycle, ZDJ is the shade intervention coefficient, and JJS is the seasonal adaptation coefficient.

[0040] Preferably, in step S4.2, the image pest recognition coefficient RXS is specifically analyzed as follows:

[0041] When RXS≤0.45, it means that there is no insect pest problem in the current image;

[0042] When RXS>0.45, it means that there is an insect pest problem in the current image.

[0043] The present application also includes a pest image recognition data processing platform, including a data acquisition module, a first analysis module, a second analysis module, a third analysis module, and a feedback module;

[0044] The data acquisition module is used to collect pest image data from different environments, different seasons, and different lighting conditions, and preprocess and reorganize the collected multi-source data to generate a first data group, a second data group, and a third data group;

[0045] The first analysis module couples the first data group, the second data group, and the third data group to generate an occlusion anomaly coefficient ZDY, performs data analysis on the occlusion anomaly coefficient ZDY, and determines whether an occlusion anomaly problem occurs based on the analysis result, and generates an occlusion intervention coefficient ZDJ based on the analysis result;

[0046] The second analysis module couples the first data group, the second data group, and the third data group to generate a seasonal adaptation coefficient JJS;

[0047] The third analysis module couples the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS of the first, second, and third data groups to generate an image pest identification coefficient RXS, performs data analysis on the image pest identification coefficient RXS, and determines whether there are pests in the image area based on the analysis results;

[0048] The feedback module feeds back various parameters to the visualization terminal. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a step diagram of the application method.

[0050] Figure 2 This is the flow chart of the application system. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] The terms "first", "second" and "third" in this application are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, a feature defined as "first", "second" and "third" may explicitly or implicitly include at least one of such features. In the description of this application, "multiple" means at least two, for example, two, three, etc., unless otherwise clearly and specifically defined. All directional indications in the embodiments of this application (such as up, down, left, right, front, back...) are only used to explain the relative positional relationship, movement, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally also include steps or units that are not listed, or may optionally also include other steps or units inherent to these processes, methods, products or devices.

[0053] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0054] Example 1: Please refer to Figure 1 , a pest image recognition data processing method, the specific steps are as follows:

[0055] S1. Data collection and enhancement: collecting pest image data from different environments, seasons, and lighting conditions, and preprocessing and reorganizing the collected multi-source data to generate the first, second, and third data sets;

[0056] S2, occlusion recovery and area estimation, coupling the first data group, the second data group, and the third data group to generate an occlusion anomaly coefficient ZDY, performing data analysis on the occlusion anomaly coefficient ZDY, and determining whether an occlusion anomaly problem occurs based on the analysis results, and generating an occlusion intervention coefficient ZDJ based on the analysis results;

[0057] S3, seasonal variation adaptability optimization, coupling the first data group, the second data group, and the third data group to generate a seasonal adaptation coefficient JJS;

[0058] S4. Multi-scale feature fusion: coupling the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS of the first, second, and third data groups to generate an image pest recognition coefficient RXS. Data analysis is performed on the image pest recognition coefficient RXS, and based on the analysis results, it is determined whether there are pests in the image area.

[0059] S5. Feedback: Feedback various parameters to the visualization terminal.

[0060] In this embodiment, in step S1, during the data collection and enhancement phase, by collecting pest image data from different environments, seasons, and lighting conditions, we ensure that the data used is highly diverse and representative. This phase involves preprocessing and reshaping the collected data to facilitate subsequent analysis. Specifically, data enhancement techniques expand the diversity of the dataset through rotation, flipping, and scaling, improving the robustness and generalization capabilities of the model. By generating the first, second, and third data sets, we provide multi-dimensional information for subsequent image recognition, helping to more comprehensively capture the characteristics of pest images.

[0061] In step S2, the system couples the first, second, and third data sets to generate an occlusion anomaly coefficient, ZDY, and performs data analysis. This step determines whether an occlusion anomaly exists in the image and generates an occlusion intervention coefficient, ZDJ, based on the analysis results. The occlusion anomaly coefficient, ZDY, helps identify potentially obscured pest areas in the image. The occlusion intervention coefficient, ZDJ, estimates the impact of occlusion within the image region and guides the system to intervene or adjust during the recognition process, restoring partially obscured pest features.

[0062] In step S3, the system couples the first, second, and third data sets to generate a seasonal adaptation coefficient, JJS. Seasonal variations affect factors such as lighting, background, and ambient temperature, directly impacting pest visibility and identification. Through seasonal adaptation optimization, the system can make specific adjustments to seasonal data to improve pest identification accuracy. This process not only helps balance data differences between seasons but also adapts to the impact of natural environmental changes, ensuring the system's continued efficient operation in a changing environment.

[0063] In step S4, the system combines the first, second, and third data sets, the occlusion intervention coefficient ZDJ, and the seasonal adaptation coefficient JJS to generate the image pest identification coefficient RXS. The key task of this process is to improve recognition accuracy by comprehensively considering information at different scales within the image through multi-scale feature fusion. By combining occlusion restoration with the seasonal adaptation coefficient JJS, the system can identify the diversity and complexity of infested areas. Finally, the system performs data analysis on the image pest identification coefficient RXS and, based on the analysis results, determines whether an infestation is present within the image area.

[0064] In step S5, the system feeds all relevant parameters to the visualization terminal in real time, enabling operators to instantly review and adjust recognition results. This step provides users with intuitive data display and interactive features for recognition results, enabling real-time monitoring and evaluation of the recognition system's operating status, providing room for optimization. This feedback mechanism enables users to make more effective data-driven decisions and interventions, improving the overall system's responsiveness and flexibility.

[0065] Traditional pest identification methods often rely on single environmental conditions or seasonal data, which can easily lead to degraded model performance in different environments. However, through data acquisition and enhanced multi-source data, this method can effectively overcome this problem and improve the system's generalization capabilities. Secondly, through occlusion recovery, regional inference, and seasonal variation adaptive optimization, the system can process occlusion and seasonal changes in images in real time, further improving recognition accuracy. Multi-scale feature fusion and data coupling effectively improve the accuracy of recognition results by comprehensively integrating different types of image information. Finally, through an instant feedback mechanism, operators can more quickly adjust and optimize recognition results, further improving the system's adaptability and efficiency.

[0066] Example 2: Please refer to Figure 1 In step S1, the specific steps of data collection and enhancement are as follows:

[0067] S1.1. Use image analysis equipment to collect image data, including obstruction area ratio, obstruction depth, number of obstructed objects, obstruction area density, obstruction area contrast, seasonal light intensity, light variation amplitude, seasonal background variation, pest growth cycle, ambient temperature index, pest area, pest contrast, edge clarity, pest boundary density, and pest obstruction degree;

[0068] S1.2, pre-process the collected parameters through the image processing system and make them dimensionless;

[0069] S1.3. Rearrange the parameters after preprocessing to generate a first data set, a second data set, and a third data set;

[0070] The first data set includes occlusion area ratio A, occlusion depth B, number of occluding objects C, occlusion area density D, and occlusion area contrast E;

[0071] The second data set includes seasonal light intensity F, light variation G, seasonal background variation H, pest growth cycle I, and ambient temperature index G;

[0072] The third data set includes the pest area K, pest contrast L, edge definition M, pest boundary density N, and pest occlusion O.

[0073] In this example, image analysis equipment collects multiple key parameters, including occlusion-related features and pest identification-related features. This significantly enriches the diversity and comprehensiveness of the dataset. Compared to traditional methods that rely solely on a few basic parameters, this multi-dimensional collection approach enables the image recognition system to comprehensively reflect multiple factors, such as environmental changes, lighting effects, and occlusion levels, thereby improving the accuracy and robustness of pest identification.

[0074] The image processing system preprocesses the multiple parameters collected and makes them dimensionless, eliminating the influence of different units and dimensions. This means that each parameter can be compared and integrated on a unified scale, effectively avoiding deviations during the calculation process. This step greatly improves the consistency and compatibility of the data, laying a solid foundation for subsequent data analysis, feature extraction, and model training.

[0075] The preprocessed parameters are reorganized and divided into the first, second, and third data groups, allowing different types of parameters to be better organized and classified in the data structure. This structured data organization method facilitates more efficient parameter analysis and feature fusion, while also enabling better specialized processing of different types of features, thereby improving the efficiency and accuracy of the model.

[0076] By extracting features from multiple dimensions and sources, the model can adapt to environmental changes such as seasonal variations, lighting conditions, and pest types, achieving enhanced generalization capabilities. In particular, the system can flexibly adjust its recognition strategy to address natural variations such as varying light intensities, occlusion conditions, and pest growth cycles, reducing errors and false positive rates.

[0077] By collecting and dimensionlessly processing multi-dimensional parameters, the adaptability and accuracy of the pest identification system are further enhanced. Compared to traditional single-data collection methods, this diversified technical approach not only improves data quality and diversity, but also provides richer and more efficient support for subsequent feature analysis and model training, effectively addressing the challenges posed by diverse environments and seasonal variations. Overall, this improved approach makes pest image recognition technology more stable under complex conditions and provides more accurate identification results.

[0078] Example 3: Please refer to Figure 1 In step S2, the specific steps of occlusion recovery and region estimation are as follows:

[0079] S2.1. Extract parameters from the first, second, and third data sets, including occlusion area ratio A, occlusion depth B, number of occluding objects C, occlusion area density D, seasonal light intensity F, seasonal background variation H, pest occlusion O, and edge definition M. Input these extracted parameters into a pre-trained deep learning framework, perform feature fusion on them through a multi-layer neural network, and calculate the occlusion anomaly coefficient ZDY.

[0080] S2.2. Perform data analysis on the occlusion anomaly coefficient ZDY to generate analysis results. Based on the analysis results, determine whether to include it in subsequent steps. The specific analysis results are as follows:

[0081] When ZDY≤0.5, it means that the occlusion anomaly coefficient ZDY does not need to be added to the subsequent calculations;

[0082] When ZDY>0.5, it means that the occlusion anomaly coefficient ZDY needs to be added to the subsequent calculations;

[0083] S2.3. Parameters are extracted from the first data group, the second data group, and the third data group, including the occlusion area ratio A, the occlusion depth B, the occlusion area density D, the seasonal light intensity F, the seasonal background change H, the pest occlusion degree O, and the pest boundary density N. The extracted multiple parameters are input into the pre-trained deep learning framework, and the features are fused through a multi-layer neural network to calculate and generate the occlusion intervention coefficient ZDJ.

[0084] In step S2.2, the occlusion anomaly coefficient ZDY is calculated using the following formula:

[0085]

[0086] Where: A is the occlusion area ratio, B is the occlusion depth, C is the number of occluding objects, D is the occlusion area density, F is the seasonal light intensity, H is the seasonal background change, O is the pest occlusion degree, M is the edge clarity, a1 and a2 are weight values, and the values of a1 and a2 are adjusted by the user.

[0087] The occlusion intervention coefficient ZDJ is calculated using the following formula:

[0088]

[0089] Where: A is the shading area ratio, B is the shading depth, D is the shading area density, F is the seasonal light intensity, H is the seasonal background change, O is the pest shading degree, N is the pest boundary density, b1 and b2 are weight values, and the values of b1 and b2 are adjusted by the user.

[0090] In this embodiment: By inputting multiple parameters from the first data group, the second data group, and the third data group into a pre-trained deep learning framework, the system can utilize a multi-layer neural network for feature fusion. This method can not only automatically extract deeper features, but also effectively capture the implicit relationship between data through the complexity of the neural network, thereby greatly improving the calculation accuracy of the occlusion anomaly coefficient ZDY and the occlusion intervention coefficient ZDJ. The introduction of deep learning enables the system to adaptively extract high-level features from different input data, avoiding the limitation of manually defining features in traditional methods, and improving the degree of automation and accuracy of image recognition. Through multi-level processing of neural networks, the system can better cope with complex data patterns and abnormal situations.

[0091] In this step, multiple parameters related to occlusion, lighting, pests, and more are comprehensively considered. Through feature fusion, the system can more comprehensively understand occlusion information and environmental changes in the image, playing a key role in the calculation of the occlusion anomaly coefficient ZDY and the occlusion intervention coefficient ZDJ. Multi-dimensional parameter fusion enables the system to comprehensively analyze possible occlusion issues in the image from multiple perspectives, and takes into account the influence of environmental factors such as seasonality and lighting, further enhancing the robustness and accuracy of recognition.

[0092] In S2.2, the system analyzes the occlusion anomaly coefficient ZDY and dynamically determines whether to include it in subsequent calculations based on the ZDY value. If ZDY ≤ 0.5, the coefficient is not required; if ZDY > 0.5, it is required. This judgment mechanism, by setting the ZDY threshold, optimizes the calculation process based on actual conditions, avoiding unnecessary complex calculations and improving computational efficiency. This dynamic decision-making mechanism makes the calculation process more flexible and can adaptively adjust the calculation model based on actual occlusion conditions, avoiding wasted computing resources when occlusion anomalies do not need to be addressed, thereby improving system efficiency and responsiveness.

[0093] By calculating the occlusion intervention coefficient (ZDJ), the system considers multiple factors, including occlusion area ratio, occlusion depth, seasonal light intensity, and pest occlusion, using a formula with adjustable weights. These weights allow users to adjust the model's sensitivity based on actual needs, further optimizing its performance. Flexible parameter adjustments allow users to tailor the model to the needs of different scenarios, thereby optimizing pest identification in specific environments. This adjustable weighting makes the method highly adaptable and capable of providing more accurate results in changing environments.

[0094] The calculation formulas for the occlusion anomaly coefficient (ZDY) and the occlusion intervention coefficient (ZDJ) combine multiple pest-related factors with environmental parameters, applying a complex weighted calculation. This comprehensive consideration enables the system to comprehensively assess the impact of occlusion in an image and take appropriate intervention measures, improving its accuracy and adaptability. This comprehensive and precise calculation method helps the system accurately determine the degree of occlusion in an image and the authenticity of infested areas, thereby enhancing the accuracy of pest identification. By integrating environmental and pest characteristics, the system can better adapt to identification tasks under diverse conditions, particularly those with significant lighting and seasonal variations.

[0095] The deep learning framework, multi-dimensional feature fusion, and dynamic judgment mechanism introduced in step S2 significantly improve the system's flexibility and accuracy in handling occlusion issues compared to traditional methods. The introduction of deep learning not only enhances the ability to recognize complex patterns but also enables the system to automatically optimize feature extraction and calculation processes, reducing the need for human intervention and configuration. Furthermore, a flexible parameter adjustment mechanism enables the model to be optimized according to the needs of different scenarios, thereby consistently providing accurate pest identification results in changing environments. Overall, these improvements enhance the system's adaptability, computational efficiency, and accuracy.

[0096] Example 4: Please refer to Figure 1In step S3, parameters are extracted from the first data group, the second data group, and the third data group, including seasonal light intensity F, seasonal background change H, ambient temperature index G, pest area K, shading area ratio A, pest shading degree O, and shading area density D. The extracted parameters are input into a pre-trained deep learning framework, and feature fusion is performed through a multi-layer neural network to calculate and generate the seasonal adaptation coefficient JJS. The specific calculation method is as follows:

[0097]

[0098] Where: F is the seasonal light intensity, H is the seasonal background change, G is the ambient temperature index, K is the area of the infested area, A is the shading area ratio, O is the degree of shading of the infested area, D is the shading area density, c1 and c2 are weight values, and b1 and b2 are user-adjustable values.

[0099] In this embodiment, by extracting multiple key parameters, including seasonal light intensity (F), seasonal background variation (H), ambient temperature index (G), pest area (K), occlusion area ratio (A), pest occlusion (O), and occlusion area density (D), the system is able to fully account for seasonal and environmental diversity. Different environmental conditions and seasonal variations directly impact pest identification, and this step, through multi-dimensional data fusion, enables the system to better adapt to these changes. This multi-dimensional parameter combination increases the system's sensitivity to environmental and seasonal variations, allowing the pest identification model to more accurately reflect pest characteristics in different seasons, reducing recognition errors caused by seasonal variations and ensuring the system maintains stability and efficiency under various natural conditions.

[0100] The extracted parameters are fed into a pre-trained deep learning framework, where they are fused using a multi-layer neural network. This process not only fully exploits the relationships between different parameters but also deeply learns the contribution of each feature to pest identification, optimizing the calculation of the seasonal adaptation coefficient (JJS). The deep learning framework automatically identifies and extracts complex patterns in the data and efficiently fuses features, enabling the system to automatically adjust and optimize identification strategies under various environmental and seasonal conditions. Compared to traditional methods that require manual feature extraction and rule-setting, the introduction of deep learning makes the system more flexible and efficient.

[0101] When calculating the seasonal adaptation coefficient JJS, adjustable weight values c1, c2, and c3 are used. These weights allow users to adjust the degree of influence of each factor on the recognition results according to actual needs. Users can flexibly set these parameters according to the requirements of different environments to maximize the performance of the system in a specific environment. The adjustability of the weights makes the system highly adaptable. Users can customize the system's response according to the actual lighting, temperature, or pest conditions to improve the accuracy of the recognition results. For example, in warm seasons, more attention may need to be paid to the ambient temperature index (G); in seasons with stronger light, the weight of seasonal light intensity (F) can be adjusted accordingly to improve recognition capabilities.

[0102] The formula incorporates both the degree of pest occlusion (O) and the occlusion area ratio (A) into the calculation, enabling the system to simultaneously account for the impact of seasonal variations and occlusion on pest identification. By comprehensively considering both occlusion and pest characteristics, the system not only adapts to seasonal changes but also effectively addresses the challenges posed by occlusion, improving recognition accuracy. This comprehensive calculation method enables the system to more accurately handle occlusion and pest issues in complex scenarios in practical applications. In particular, in environments with significant obstructions, the system maintains high recognition accuracy, ensuring that pests are not missed due to partial occlusion or environmental changes.

[0103] By calculating the seasonal adaptation coefficient (JJS), the system can better adapt to diverse environmental changes, particularly seasonal variations in light, temperature, and background. Compared to traditional methods that rely solely on fixed models or a single data source, this approach significantly enhances the system's generalization capabilities. The seasonal adaptation coefficient (JJS) enables the system to consistently provide highly accurate pest identification results across different seasons and environments, reducing recognition errors caused by seasonal variations. This allows the system to maintain strong performance in changing environments, making it more flexible and reliable in practical applications.

[0104] Example 5: Please refer to Figure 1 In step S4, the specific steps of multi-scale feature fusion are as follows:

[0105] S4.1. Extract parameters from the first, second, and third data sets, including the obscured area ratio A, obscured area contrast E, seasonal light intensity F, pest area K, and pest growth period I. Input these extracted parameters, along with the obscured intervention coefficient ZDJ and the seasonal adaptation coefficient JJS, into a pre-trained deep learning framework. Feature fusion is performed using a multi-layer neural network to calculate the image pest recognition coefficient RXS.

[0106] S4.2. Perform data analysis on the image pest identification coefficient RXS to generate analysis results. Based on the analysis results, determine whether to add it to the subsequent steps.

[0107] In step S4.1, the specific calculation formulas for the image pest recognition coefficient RXS are as follows:

[0108]

[0109] Where A is the shaded area ratio, E is the shaded area contrast, F is the seasonal light intensity, K is the area of the pest area, I is the pest growth cycle, ZDJ is the shade intervention coefficient, and JJS is the seasonal adaptation coefficient.

[0110] In step S4.2, the specific analysis method for the image pest recognition coefficient RXS is as follows:

[0111] When RXS≤0.45, it means that there is no insect pest problem in the current image;

[0112] When RXS>0.45, it means there is an insect pest problem in the current image.

[0113] In this embodiment: In S4.1, by extracting multiple important parameters such as occlusion area ratio (A), occlusion area contrast (E), seasonal light intensity (F), pest area (K) and pest growth cycle (I), and combining the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS, the system can achieve multi-scale feature fusion. In this way, the system can not only perform analysis at a single level, but also integrate data in multiple dimensions to provide a more comprehensive feature expression for subsequent pest identification. Multi-scale feature fusion enhances the system's ability to process complex information in the image, and can simultaneously consider different types of features and environmental factors. By combining multiple scales, the system can provide more accurate pest identification results under the influence of complex environmental changes, different seasons, and obstructions, thereby improving the robustness and accuracy of the system.

[0114] The extracted parameters, along with the occlusion intervention coefficient ZDJ and seasonal adaptation coefficient JJS, are input into a pre-trained deep learning framework, where the different features are fused through a multi-layer neural network. This allows the system to learn hidden patterns and complex relationships based on large amounts of data, rather than relying on the design of manual features. This automated learning capability greatly improves the accuracy and efficiency of the recognition process. Through automatic feature fusion, the deep learning framework can not only extract high-level feature information, but also efficiently identify imperceptible details in the image, compensating for the limitations of traditional methods in complex scenarios. Compared to manually designed features, deep learning can better cope with complex patterns in changing environments, improving the system's adaptability and performance.

[0115] In S4.1, the formula for calculating the image pest recognition coefficient (RXS) combines multiple parameters and coefficients to accurately reflect the impact of different factors on pest identification. Through this comprehensive calculation, the system can comprehensively consider the interactions between occlusion, seasonal changes, and pest characteristics, thereby generating a more accurate recognition coefficient. This optimized calculation formula can carefully reflect the impact of multiple factors on pest image recognition, and through the weighted calculation, the image pest recognition coefficient (RXS) is calculated, making the final judgment more precise. In complex environments, this flexible calculation method can effectively handle occlusion, lighting changes, and the diversity of pest areas, improving the accuracy of pest identification.

[0116] In S4.2, the system automatically determines the presence of pests by analyzing the image's pest identification coefficient (RXS). When RXS ≤ 0.45, there is no pest problem within the image; when RXS > 0.45, there is a pest problem. This decision-making mechanism automatically determines the presence of pests by setting a threshold, eliminating the need for human intervention. This automated decision-making mechanism reduces the need for human intervention and improves the efficiency and accuracy of the entire system. Through quantitative threshold judgment, the system can quickly identify the presence of pests in images, significantly improving processing speed and real-time performance. This is particularly effective in large-scale pest monitoring applications that require rapid response.

[0117] By combining the occlusion intervention coefficient (ZDJ) and the seasonal adaptation coefficient (JJS), the system can optimize based on specific environmental conditions, further improving the accuracy of image pest recognition. Flexible adjustment of these coefficients enables the system to automatically optimize the recognition algorithm in a variety of environments. Flexible parameter adjustment allows the system to optimize according to the needs of different scenarios. Users can customize adjustments based on environmental changes and pest types, thereby improving recognition adaptability and accuracy. The system maintains efficient and accurate recognition performance despite varying seasons, lighting, and occlusion conditions.

[0118] By combining multi-scale feature fusion with a deep learning framework, step S4 significantly improves the accuracy and adaptability of pest image recognition. Compared to traditional methods, this system, which integrates multiple feature dimensions, automated decision-making, and flexible adjustments, can provide more accurate pest identification results in different environments and conditions. The automated calculation of the image pest recognition coefficient (RXS) and the threshold-based decision-making mechanism make the entire recognition process more efficient and real-time. This optimization enables the system to flexibly respond to complex environmental changes, improving the overall performance and practicality of pest identification.

[0119] This application also includes a pest image recognition data processing platform, see Figure 2 , including a data acquisition module, a first analysis module, a second analysis module, a third analysis module and a feedback module;

[0120] The data acquisition module is used to collect pest image data from different environments, different seasons, and different lighting conditions, and preprocess and reorganize the collected multi-source data to generate a first data group, a second data group, and a third data group;

[0121] The first analysis module couples the first data group, the second data group, and the third data group to generate an occlusion anomaly coefficient ZDY, performs data analysis on the occlusion anomaly coefficient ZDY, and determines whether an occlusion anomaly problem occurs based on the analysis result, and generates an occlusion intervention coefficient ZDJ based on the analysis result;

[0122] The second analysis module couples the first data group, the second data group, and the third data group to generate a seasonal adaptation coefficient JJS;

[0123] The third analysis module couples the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS of the first, second, and third data groups to generate an image pest identification coefficient RXS, performs data analysis on the image pest identification coefficient RXS, and determines whether there are pests in the image area based on the analysis results;

[0124] The feedback module feeds back various parameters to the visualization terminal.

[0125] In addition, the functional units in the various embodiments of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

[0126] The above detailed description of the specific embodiments of the invention is intended only as an example, and the present application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions of the invention are also within the scope of the present application. Therefore, equivalent changes, modifications, and improvements made without departing from the spirit and scope of the present application should be included within the scope of the present application.

Claims

1. A method for processing pest image recognition data, characterized in that: The specific steps are as follows: S1. Data collection and enhancement: collecting pest image data from different environments, seasons, and lighting conditions, and preprocessing and reorganizing the collected multi-source data to generate the first, second, and third data sets; S2, occlusion recovery and area estimation, coupling the first data group, the second data group, and the third data group to generate an occlusion anomaly coefficient ZDY, performing data analysis on the occlusion anomaly coefficient ZDY, and determining whether an occlusion anomaly problem occurs based on the analysis results, and generating an occlusion intervention coefficient ZDJ based on the analysis results; S3, seasonal variation adaptability optimization, coupling the first data group, the second data group, and the third data group to generate a seasonal adaptation coefficient JJS; S4. Multi-scale feature fusion: coupling the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS of the first, second, and third data groups to generate an image pest recognition coefficient RXS. Data analysis is performed on the image pest recognition coefficient RXS, and based on the analysis results, it is determined whether there are pests in the image area. S5. Feedback: Feedback various parameters to the visualization terminal.

2. The pest image recognition data processing method according to claim 1, characterized in that: In step S1, the specific steps of data collection and enhancement are as follows: S1.

1. Use image analysis equipment to collect image data, including obstruction area ratio, obstruction depth, number of obstructed objects, obstruction area density, obstruction area contrast, seasonal light intensity, light variation amplitude, seasonal background variation, pest growth cycle, ambient temperature index, pest area, pest contrast, edge clarity, pest boundary density, and pest obstruction degree; S1.2, pre-process the collected parameters through the image processing system and make them dimensionless; S1.

3. Rearranging the pre-processed parameters to generate a first data set, a second data set, and a third data set; The first data set includes occlusion area ratio A, occlusion depth B, number of occluding objects C, occlusion area density D, and occlusion area contrast E; The second data set includes seasonal light intensity F, light variation G, seasonal background variation H, pest growth cycle I, and ambient temperature index G; The third data set includes the pest area K, pest contrast L, edge definition M, pest boundary density N, and pest occlusion O.

3. The pest image recognition data processing method according to claim 2, characterized in that: In step S2, the specific steps of occlusion recovery and area estimation are as follows: S2.

1. Extract parameters from the first, second, and third data sets, including occlusion area ratio A, occlusion depth B, number of occluding objects C, occlusion area density D, seasonal light intensity F, seasonal background variation H, pest occlusion O, and edge definition M. Input these extracted parameters into a pre-trained deep learning framework, perform feature fusion on them through a multi-layer neural network, and calculate the occlusion anomaly coefficient ZDY. S2.

2. Perform data analysis on the occlusion anomaly coefficient ZDY to generate analysis results. Based on the analysis results, determine whether to include it in subsequent steps. The specific analysis results are as follows: When ZDY≤0.5, it means that the occlusion anomaly coefficient ZDY does not need to be added to the subsequent calculations; When ZDY>0.5, it means that the occlusion anomaly coefficient ZDY needs to be added to the subsequent calculations; S2.

3. Parameters are extracted from the first data group, the second data group, and the third data group, including the occlusion area ratio A, the occlusion depth B, the occlusion area density D, the seasonal light intensity F, the seasonal background change H, the pest occlusion degree O, and the pest boundary density N. The extracted multiple parameters are input into the pre-trained deep learning framework, and the features are fused through a multi-layer neural network to calculate and generate the occlusion intervention coefficient ZDJ.

4. The pest image recognition data processing method according to claim 3, characterized in that: In step S2.2, the occlusion anomaly coefficient ZDY is calculated using the following formula: Where: A is the occlusion area ratio, B is the occlusion depth, C is the number of occluding objects, D is the occlusion area density, F is the seasonal light intensity, H is the seasonal background change, O is the pest occlusion degree, M is the edge clarity, a1 and a2 are weight values, and the values of a1 and a2 are adjusted by the user.

5. The pest image recognition data processing method according to claim 4, characterized in that: In step S2.3, the occlusion intervention coefficient ZDJ is calculated using the following formula: Where: A is the shading area ratio, B is the shading depth, D is the shading area density, F is the seasonal light intensity, H is the seasonal background change, O is the pest shading degree, N is the pest boundary density, b1 and b2 are weight values, and the values of b1 and b2 are adjusted by the user.

6. The pest image recognition data processing method according to claim 5, characterized in that: In step S3, parameters are extracted from the first, second, and third data groups, including seasonal light intensity F, seasonal background change H, ambient temperature index G, pest area K, occlusion area ratio A, pest occlusion degree O, and occlusion area density D. The extracted parameters are input into a pre-trained deep learning framework, and feature fusion is performed through a multi-layer neural network to calculate and generate the seasonal adaptation coefficient JJS. The specific calculation method is as follows: Where: F is the seasonal light intensity, H is the seasonal background change, G is the ambient temperature index, K is the area of the pest area, A is the shading area ratio, O is the pest shading degree, D is the shading area density, c1 and c2 are weight values, and the values of b1 and b2 are adjusted by the user.

7. The pest image recognition data processing method according to claim 6, characterized in that: In step S4, the specific steps of multi-scale feature fusion are as follows: S4.

1. Extract parameters from the first, second, and third data sets, including the obscured area ratio A, obscured area contrast E, seasonal light intensity F, pest area K, and pest growth period I. Input these extracted parameters, along with the obscured intervention coefficient ZDJ and the seasonal adaptation coefficient JJS, into a pre-trained deep learning framework. Feature fusion is performed using a multi-layer neural network to calculate the image pest recognition coefficient RXS. S4.

2. Perform data analysis on the image pest identification coefficient RXS to generate an analysis result. Based on the analysis result, determine whether there is a pest problem in the current image area.

8. The pest image recognition data processing method according to claim 7, characterized in that: In step S4.1, the specific calculation formulas for the image pest recognition coefficient RXS are as follows: Where A is the shaded area ratio, E is the shaded area contrast, F is the seasonal light intensity, K is the area of the pest area, I is the pest growth cycle, ZDJ is the shade intervention coefficient, and JJS is the seasonal adaptation coefficient.

9. The pest image recognition data processing method according to claim 8, characterized in that: In step S4.2, the specific analysis method for the image pest recognition coefficient RXS is as follows: When RXS≤0.45, it means that there is no insect pest problem in the current image; When RXS>0.45, it means that there is an insect pest problem in the current image.

10. A pest image recognition data processing platform, characterized in that: The pest image recognition data processing platform is used to execute the pest image recognition data processing method according to any one of claims 1 to 9, comprising a data acquisition module, a first analysis module, a second analysis module, a third analysis module, and a feedback module; The data acquisition module is used to collect pest image data from different environments, different seasons, and different lighting conditions, and preprocess and reorganize the collected multi-source data to generate a first data group, a second data group, and a third data group; The first analysis module couples the first data group, the second data group, and the third data group to generate an occlusion anomaly coefficient ZDY, performs data analysis on the occlusion anomaly coefficient ZDY, and determines whether an occlusion anomaly problem occurs based on the analysis result, and generates an occlusion intervention coefficient ZDJ based on the analysis result; The second analysis module couples the first data group, the second data group, and the third data group to generate a seasonal adaptation coefficient JJS; The third analysis module couples the occlusion intervention coefficient ZDJ and the seasonal adaptation coefficient JJS of the first, second, and third data groups to generate an image pest identification coefficient RXS, performs data analysis on the image pest identification coefficient RXS, and determines whether there are pests in the image area based on the analysis results; The feedback module feeds back various parameters to the visualization terminal.

Images