A method and system for identifying crop pests based on visual recognition

Abstract

This invention discloses a method and system for identifying crop pests based on visual recognition, belonging to the field of pest identification technology. A crop pest identification system based on visual recognition includes a crop visual processing module and a crop pest identification module. This invention constructs a multi-source information collaborative judgment mechanism by integrating visual image features, volatile gas concentrations, and environmental parameters, effectively distinguishing between physiological abnormalities and pathogenic diseases, reducing misjudgments and missed judgments caused by environmental interference, and significantly improving the accuracy and reliability of identification results. By combining gas sensor data to analyze crop respiration characteristics and maturity status, it can identify crop physiological abnormalities before visual symptoms appear, achieving early detection and warning of pests. Dynamically adjusting the identification threshold and sensitivity according to different growth stages supports targeted prevention and control decisions, improving the timeliness of prevention and control.

Description

Technical Field

[0001] This invention relates to the field of pest identification technology, and in particular to a method and system for identifying crop pests based on visual recognition. Background Technology

[0002] In existing technologies, vision-based methods for identifying crop diseases and pests mainly rely on image processing and deep learning models. Although they have achieved automated identification to some extent, they still have obvious limitations: First, relying solely on image features makes it difficult to distinguish between physiological diseases and infectious diseases, and they are easily affected by environmental factors such as light and shading, leading to misjudgments and missed judgments. Second, they lack dynamic perception of crop growth status and cannot combine physiological indicators such as maturity and respiration characteristics for comprehensive judgment, resulting in insufficient accuracy and timeliness of identification. Third, they fail to effectively integrate biochemical information such as volatile organic compounds, making it difficult to achieve early warning and precise control of pests.

[0003] Therefore, there is a need to provide a method and system for identifying crop pests based on the fusion of vision and gas sensing. Summary of the Invention

[0004] This invention aims to provide a method and system for identifying crop pests based on visual recognition, which can dynamically adjust the pest identification sensitivity and threshold to achieve accurate identification and early warning of pest types at different growth stages and under different environmental conditions, significantly improving the identification accuracy.

[0005] A method for identifying crop pests based on visual recognition includes the following steps:

[0006] Real-time acquisition of crop image data using vision devices yields crop surface image data; a crop visual recognition model is constructed; and the crop surface image data is processed based on the crop visual recognition model to obtain abnormal crop surface images and characteristics of crop growth stages.

[0007] The concentration of volatile gases around crops is collected synchronously using gas sensing devices; the crop status is identified based on the volatile gas concentration to obtain crop maturity characteristics; the crop maturity characteristics are used to represent the maturity stage and expected maturity time of crops.

[0008] Obtain crop environmental characteristics; identify crop pests based on crop environmental characteristics, crop maturity characteristics, and abnormal crop surface images, and obtain crop pest identification results.

[0009] As a preferred embodiment of the present invention, the specific steps for processing crop surface image data based on a crop visual recognition model include:

[0010] The crop visual recognition model includes an image segmentation module, an anomaly feature enhancement module, and a growth feature judgment module;

[0011] In the image segmentation module, the crop surface image data is segmented by abnormal features, and the crop abnormal image containing abnormal features and the crop background image containing a large background are extracted. In the abnormal feature enhancement module, the crop abnormal image is enhanced to obtain the crop abnormal surface image.

[0012] In the growth feature judgment module, the crop growth stage is identified in the background image of the crop to be identified, and the crop growth stage features are obtained. Specifically, in the growth feature judgment module, multi-scale convolutional feature extraction is used to extract features layer by layer from the crop region of the background image of the crop to be identified, and the crop growth features are obtained. The color information in the crop growth features is extracted to obtain the crop growth color features. The trained classifier is used to judge the crop growth color features to obtain the crop growth stage features.

[0013] As a preferred embodiment of the present invention, the specific steps for constructing the abnormal feature enhancement module in the crop visual recognition model include:

[0014] The YOLOv8 model is used as the base model for the anomaly feature enhancement module in the crop visual recognition model. In the backbone network of the base model, ECA attention mechanism is introduced as a convolutional feature extraction unit, a multi-task learning branch is added to the head part as a crop feature classification unit, and CSPNet structure is used as a crop feature fusion unit in the neck part. At the same time, the Alpha-IoU function is set as the loss function to obtain the optimized anomaly feature enhancement module.

[0015] The abnormal feature enhancement module processes the crop image to be identified as abnormal to obtain an abnormal surface image of the crop.

[0016] As a preferred embodiment of the present invention, the specific steps for identifying crop status based on volatile gas concentration include:

[0017] Data processing based on volatile gas concentrations yields standard time-series gas concentration data; the mean gas concentration of the standard time-series gas concentration data is compared with the corresponding volatile gas threshold range of crops to obtain the basic maturity level of crops.

[0018] Meanwhile, the concentrations of the first and second gases are extracted from the standard time-series gas concentration data and compared to obtain the active gas ratio. When the active gas ratio exceeds the preset gas ratio, the corresponding crop is judged to be in the respiratory climax period; otherwise, the corresponding crop is in the respiratory plateau period. At the same time, the diurnal variation pattern of the third gas concentration is analyzed to obtain the volatile substance release characteristics.

[0019] A gas concentration change trend model is established based on the respiratory climax or respiratory stationary phase. By continuously monitoring the slope of the changes in the first, second, and third gas concentrations, the characteristics of the predicted crop ripening process can be obtained.

[0020] The crop maturity characteristics are obtained by fusing the basic maturity level of crops, the characteristics of volatile substance release, and the characteristics of predicted crop maturity process.

[0021] As a preferred embodiment of the present invention, the specific steps for identifying crop pests based on crop environmental characteristics, crop maturity characteristics, and abnormal crop surface images include:

[0022] Correlation analysis is performed based on crop environmental characteristics and crop maturity characteristics to obtain external characteristics of crop diseases; the identification sensitivity of different types of pests is matched according to the external characteristics of crop diseases.

[0023] The pest identification threshold for abnormal crop surface images is adjusted based on the identification sensitivity to obtain the corrected pest distribution. Spatial correlation analysis is performed on crop environmental characteristics and the corrected pest distribution, and crop pests are identified in combination with crop maturity characteristics to obtain the crop pest identification results.

[0024] As a preferred embodiment of the present invention, for different varieties of crops, the first gas, the second gas, and the third gas represent different types of gases.

[0025] A visual recognition-based crop pest identification system includes:

[0026] The crop visual processing module includes a visual processing unit and an auxiliary processing unit. The visual processing unit is used to acquire crop image data in real time based on visual devices to obtain crop surface image data; construct a crop visual recognition model; and process the crop surface image data based on the crop visual recognition model to obtain crop abnormal surface images and crop growth stage characteristics.

[0027] The auxiliary processing unit is used to synchronously collect the concentration of volatile gases around crops based on gas sensing devices; to identify the crop status based on the volatile gas concentration, and to obtain the crop maturity characteristics; the crop maturity characteristics are used to represent the maturity stage and expected maturity time of the crop.

[0028] The crop pest identification module includes a pest identification unit; the pest identification unit is used to acquire crop environmental characteristics; based on crop environmental characteristics, crop maturity characteristics and abnormal crop surface images, pest identification is performed on the crop to obtain crop pest identification results.

[0029] The present invention has the following advantages:

[0030] 1. This invention constructs a multi-source information collaborative judgment mechanism by integrating visual image features, volatile gas concentrations, and environmental parameters. This mechanism effectively distinguishes between physiological abnormalities and pathogenic diseases, reduces misjudgments and omissions caused by environmental interference, and significantly improves the accuracy and reliability of identification results. By combining gas sensor data to analyze crop respiration characteristics and maturity status, it can identify crop physiological abnormalities before visual symptoms appear, enabling early detection and warning of pests. The recognition threshold and sensitivity are dynamically adjusted according to different growth stages to support targeted prevention and control decisions and improve the timeliness of prevention and control.

[0031] 2. This invention, by constructing a multi-module collaborative visual processing model and a gas-environment correlation analysis mechanism, can adapt to different crop varieties, growth stages, and complex field environments, possessing good generalization ability and scenario adaptability, and is suitable for large-scale, multi-variety agricultural application scenarios; by dynamically adjusting the pest identification threshold based on pest identification results and crop growth stages, it provides precise control timing and appropriate control measures, effectively reducing the waste of agricultural resources and improving crop yield and quality. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of a crop pest identification system based on visual recognition used in an embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0034] Example 1: A method for identifying crop pests based on visual recognition, comprising the following steps:

[0035] Real-time acquisition of crop image data using vision devices yields crop surface image data; a crop visual recognition model is constructed; and the crop surface image data is processed based on the crop visual recognition model to obtain abnormal crop surface images and characteristics of crop growth stages.

[0036] Vision devices such as high-definition cameras, drones, or ground robots installed in farmland or orchards collect real-time image data of different parts of crops, including but not limited to leaves, fruits, and branches. These images are captured under varying lighting and environmental conditions to ensure clear and accurate images of the crop surface under diverse external environments. This image data forms the basis for subsequent identification and analysis, providing information such as crop growth status and surface damage, and offering direct input for subsequent pest identification and growth stage analysis.

[0037] The specific steps for processing crop surface image data based on a crop visual recognition model include:

[0038] The crop visual recognition model includes an image segmentation module, an anomaly feature enhancement module, and a growth feature judgment module;

[0039] In the image segmentation module, the crop surface image data is segmented by abnormal features, and the crop abnormal image containing abnormal features and the crop background image containing a large background are extracted. In the abnormal feature enhancement module, the crop abnormal image is enhanced to obtain the crop abnormal surface image.

[0040] The image segmentation module is tasked with classifying and segmenting crop surface image data, extracting regions with obvious abnormal features from large background areas. Through image segmentation technology, the module can effectively distinguish areas containing abnormal phenomena such as pests, lesions, and yellowing; these areas are marked as abnormal images to be identified. Meanwhile, the background area includes healthy parts of the crop and a large area of ​​the surrounding environment, serving as a background image for subsequent analysis. By effectively segmenting the image, the module can separate the abnormal areas requiring focused processing from the simpler background areas, reducing the complexity of subsequent processing and ensuring accurate identification of pests and diseases.

[0041] In the growth feature judgment module, the crop growth stage is identified in the background image of the crop to be identified, and the crop growth stage features are obtained. Specifically, in the growth feature judgment module, multi-scale convolutional feature extraction is used to extract features layer by layer from the crop region of the background image of the crop to be identified, and the crop growth features are obtained. The color information in the crop growth features is extracted to obtain the crop growth color features. The trained classifier is used to judge the crop growth color features to obtain the crop growth stage features.

[0042] In the growth feature judgment module, the growth stage of the crop is identified in the background image to be recognized. The core task of the growth feature judgment module is to extract the growth stage features of the crop from the background image. First, through the multi-scale convolution feature extraction method, the module extracts different levels of features of the crop area from the background image layer by layer. These features include the size and shape changes of the crop leaves, fruit enlargement, color changes, etc., which can reflect the changes of the crop in different growth stages. Then, by extracting the chromaticity information of the crop, the growth chromaticity features of the crop are further obtained. These chromaticity features can help the system determine the growth stage of the crop, such as the transition from the vegetative growth stage to the fruit enlargement stage.

[0043] Using a trained classifier, the growth feature judgment module judges the extracted growth color features. The classifier classifies crop images into different growth stages, such as the initial stage, swelling stage, and maturity stage, by comparing them with known growth stage data. Based on the operation of the classifier, the system can accurately identify the current growth stage of the crop, providing a basis for subsequent pest prediction and control measures. The above process not only improves the accuracy of crop growth stage judgment, but also dynamically reflects the changes of crops in different growth stages, further optimizing the accuracy of pest identification.

[0044] The specific steps for constructing an anomaly feature enhancement module in a crop visual recognition model include:

[0045] The YOLOv8 model is used as the base model for the anomaly feature enhancement module in the crop visual recognition model. In the backbone network of the base model, ECA attention mechanism is introduced as a convolutional feature extraction unit, a multi-task learning branch is added to the head part as a crop feature classification unit, and CSPNet structure is used as a crop feature fusion unit in the neck part. At the same time, the Alpha-IoU function is set as the loss function to obtain the optimized anomaly feature enhancement module.

[0046] When constructing the anomaly feature enhancement module in the crop visual recognition model, we first use the YOLOv8 model as the base model. YOLOv8 is an object detection model that can achieve fast and accurate object recognition. On this base model, we make further improvements to enhance the model's performance in processing anomalies in crop surface images. To enhance the model's ability to perceive subtle features, we introduce the ECA attention mechanism into the YOLOv8 backbone network. The ECA attention mechanism can dynamically adjust the weights of each channel in the convolutional layer, allowing the model to focus more on regions with stronger features for crop anomalies (such as pests, lesions, etc.). Through this mechanism, the network can effectively improve its sensitivity to anomalies when extracting image features, thereby strengthening the recognition of subtle anomalies.

[0047] In the Head section of YOLOv8, a multi-task learning branch was added to the base model. This branch can perform multiple tasks simultaneously, including object detection and crop feature classification. This allows the model to not only identify abnormal areas on the surface of crops but also to further classify these abnormal areas, such as distinguishing different types of pests or diseases. Multi-task learning not only improves the overall performance of the model but also allows it to handle more tasks within the same network, thereby enhancing the efficiency and accuracy of identification.

[0048] In the Neck section, the CSPNet structure was used for feature fusion. The CSPNet structure can effectively fuse features from different levels. By separating and merging feature maps, the model can extract features at multiple scales simultaneously, thereby improving the detection capability of complex scenes and abnormal features at different scales. Through this structure, various features in crop surface images (such as small insect pests, lesions and their morphological changes) can be better fused, enhancing the overall feature representation capability.

[0049] The Alpha-IoU function was set as the loss function for this model. The Alpha-IoU function is an optimized intersection-union ratio loss function that can dynamically adjust the weights of different regions, providing higher accuracy in the localization and classification of targets. By using the Alpha-IoU loss function, the model can more accurately determine the location of abnormal areas in crops, while reducing attention to background areas and improving the recognition effect of abnormal surface images of crops.

[0050] In the anomaly feature enhancement module, the crop image to be identified is processed to obtain an abnormal crop surface image. In the optimized anomaly feature enhancement module, the crop image to be identified is processed to obtain an abnormal crop surface image with more prominent abnormal features. The enhanced image can clearly reflect the abnormal conditions such as diseases and pests on the crop surface, providing more accurate input data for subsequent pest identification and further improving the accuracy of pest detection and classification.

[0051] In the anomaly feature enhancement module, feature enhancement processing is performed on the abnormal images to be identified. The core of the anomaly feature enhancement module is to enhance the subtle abnormal features in the image, making them more obvious and facilitating further analysis. By using image enhancement algorithms, the system can magnify and display features such as insect traces, insect holes, or lesions in abnormal images, making them more distinct from normal crop areas. The enhanced abnormal surface images will better reflect the signs of crop diseases or pests, thereby providing clearer and more accurate visual data for subsequent pest identification, ensuring accurate location and classification of pests.

[0052] The concentration of volatile gases around crops is collected synchronously using gas sensing devices; the crop status is identified based on the volatile gas concentration to obtain crop maturity characteristics; the crop maturity characteristics are used to represent the maturity stage and expected maturity time of crops.

[0053] The specific steps for identifying crop status based on volatile gas concentrations include:

[0054] Data processing based on volatile gas concentrations yields standard time-series gas concentration data; the mean gas concentration of the standard time-series gas concentration data is compared with the corresponding volatile gas threshold range of crops to obtain the basic maturity level of crops.

[0055] Meanwhile, the concentrations of the first and second gases are extracted from the standard time-series gas concentration data and compared to obtain the active gas ratio. When the active gas ratio exceeds the preset gas ratio, the corresponding crop is judged to be in the respiratory climax period; otherwise, the corresponding crop is in the respiratory plateau period. At the same time, the diurnal variation pattern of the third gas concentration is analyzed to obtain the volatile substance release characteristics.

[0056] A gas concentration change trend model is established based on the respiratory climax or respiratory stationary phase. By continuously monitoring the slope of the changes in the first, second, and third gas concentrations, the characteristics of the predicted crop ripening process can be obtained.

[0057] The characteristics of crop maturity are obtained by fusing the basic maturity level of crops, the characteristics of volatile substance release, and the characteristics of predicted crop maturity process.

[0058] For different varieties of crops, the first gas, the second gas, and the third gas represent different types of gases.

[0059] In the process of identifying crop status by measuring volatile gas concentrations, the first step is to collect real-time data on the volatile gas concentrations around the crops using gas sensing devices. This data includes concentration information for different gases, which exhibit different concentration characteristics at different growth stages of the crops. For example, for crops like citrus, the gas sensors detect gases including ethyl acetate (first gas), ethanol (second gas), and ethylene (third gas). Ethyl acetate and ethanol are typically associated with fruit ripening, while ethylene is an important plant hormone that usually increases significantly during fruit ripening, especially during the critical stages of fruit enlargement and ripening.

[0060] Gas sensing devices are used to monitor the concentration of volatile organic compounds (VOCs) around crops. These devices have wide applications in agriculture, particularly in monitoring crop maturity, early warning of pests and diseases, and environmental monitoring. Examples include electronic noses, gas sensor arrays, and gas chromatographs. An electronic nose is a sensing device that mimics human olfaction and is widely used to detect gases and VOCs in the air. Gas sensor arrays include multiple different types of gas sensors and are typically used to accurately detect various gases, such as ammonia, ethylene, ethanol, propylene, and benzene. Gas chromatographs can separate and quantify gases in air samples and are suitable for detailed analysis of various VOCs in the air around crops. Especially in agricultural research, gas chromatographs can provide high-resolution gas composition analysis data. However, gas chromatographs are typically used in laboratories and are not well-suited for outdoor use in practice.

[0061] Next, the collected gas concentration data is processed to obtain standard time-series gas concentration data. This process includes preprocessing the gas concentration data, such as denoising and standardization, to ensure data quality. The processed data is then compared with the mean gas concentration to determine the basic maturity level of the crop. Taking citrus as an example, when the concentrations of ethyl acetate and ethanol reach specific thresholds, it indicates that the fruit has entered a certain stage of ripening, thus determining the maturity level of the crop.

[0062] Following this, we compared the first gas (ethyl acetate) and the second gas (ethanol) in the standard time-series gas concentration data to obtain the active gas ratio, which reflects the release of these gases during fruit ripening. If the active gas ratio exceeds the preset gas ratio threshold, it means that the citrus fruit is in the respiratory climacteric phase, that is, the fruit is in the rapid ripening process and the gas concentration changes are relatively drastic; otherwise, the citrus fruit is in the respiratory stationary phase, its ripening process is relatively stable, and the gas concentration changes are small. Further analysis of the diurnal variation pattern of the third gas concentration (such as ethylene) can reveal the release characteristics of this gas. Ethylene usually increases rapidly during the fruit ripening period and is an important indicator of maturity.

[0063] Based on the changes in the gas concentration data mentioned above, we can establish a gas concentration change trend model. By continuously monitoring the slope of the concentration changes of the first gas (ethyl acetate), the second gas (ethanol), and the third gas (ethylene), we can predict the characteristics of the ripening process of citrus. For example, if the concentration of ethylene rises rapidly over several consecutive days, and the concentrations of ethyl acetate and ethanol also show a similar growth trend, then we can infer that the citrus is about to enter the full ripening stage, and the expected harvest time will be clear.

[0064] Based on the basic maturity level of crops, the characteristics of volatile substance release, and the predicted maturity process characteristics, this information is fused to obtain the final crop maturity characteristics. For citrus, crop maturity characteristics can help determine the fruit's maturity stage and the expected maturity time, thus providing a precise basis for pest early warning and harvest time. Through this process, combined with the dynamic changes in gas concentration and the physiological state of crops, the maturity of citrus can be judged more accurately, providing data support for precise crop management and pest control.

[0065] Specifically, in crop pest identification, the setting of volatile gas threshold ranges and preset gas ratios is based on the physiological characteristics of crops and their natural relationship with environmental changes, rather than values ​​set arbitrarily by humans. These thresholds are set in accordance with natural laws and are determined through a large amount of experimental data, gas concentration change patterns and crop maturation processes, so as to ensure that the identification model is closely integrated with the dynamic process of actual crop growth and pest occurrence.

[0066] Obtain crop environmental characteristics; identify crop pests based on crop environmental characteristics, crop maturity characteristics, and abnormal crop surface images, and obtain crop pest identification results;

[0067] The specific steps for identifying crop pests based on crop environmental characteristics, crop maturity characteristics, and abnormal crop surface images include:

[0068] Correlation analysis is performed based on crop environmental characteristics and crop maturity characteristics to obtain external characteristics of crop diseases; the identification sensitivity of different types of pests is matched according to the external characteristics of crop diseases.

[0069] After acquiring the environmental characteristics of crops, a correlation analysis is conducted based on these characteristics and maturity features. By analyzing factors such as temperature, humidity, light intensity, and soil moisture in the crop's environment, combined with maturity characteristics (such as fruit enlargement and color changes), the current growth status of the crop and its relationship with the environment can be understood. The correlation analysis between environmental and maturity features helps reveal the sensitivity of crops to environmental changes at different growth stages, further providing a basis for the potential risk of pest outbreaks. For example, certain pests and diseases may be more likely to occur in warm and humid environments, while crops are more susceptible to specific pests during the maturity stage. This analysis establishes a potential correlation between environmental factors and pests, providing more accurate data for subsequent precise pest prediction.

[0070] This method matches the external characteristics of crop diseases with the identification sensitivity of different types of pests. External characteristics include indicative changes on the crop surface such as lesions, yellowing, and insect holes, which are correlated with pest types. Based on these external characteristics and existing pest data, the identification sensitivity of corresponding pests is determined. For example, spots or discoloration on the leaves of some crops may indicate aphids or whiteflies, while cracks on the fruit surface may be related to fruit borers. By adjusting the identification sensitivity based on changes in environmental and external disease characteristics, the sensitivity for identifying specific pest types is improved.

[0071] The pest identification threshold for abnormal crop surface images is adjusted based on the identification sensitivity to obtain the corrected pest distribution. The crop environmental characteristics and the corrected pest distribution are spatially correlated and combined with the crop maturity characteristics to identify crop pests, thus obtaining the crop pest identification results.

[0072] Based on the sensitivity of the identification, the pest identification threshold for abnormal surface images of crops is corrected. By dynamically adjusting the pest identification threshold, adaptive adjustments can be made to meet the pest identification needs of different environments, different crop varieties, and different maturity stages. After the pest identification threshold is corrected, a more accurate pest distribution map will be obtained, thereby reducing the phenomenon of false identification and missed identification. The corrected pest distribution map can reflect the pest areas that may exist in certain specific areas or growth stages of crops, which helps to implement subsequent control measures.

[0073] Spatial correlation analysis is performed on crop environmental characteristics and corrected pest distribution. This analysis identifies the distribution patterns of pests in farmland or orchards and combines this with crop maturity characteristics (such as maturity level and fruit enlargement) for comprehensive evaluation. Crop maturity significantly impacts pest occurrence; for example, the risk of pest outbreaks increases as fruit matures. By combining spatial data on environmental characteristics, corrected pest distribution, and maturity characteristics, comprehensive pest identification is performed, yielding accurate results. Specifically, the pest identification results indicate the presence of pests on or around the crop surface, along with the type, location, severity, and potential impact area, based on multi-dimensional information such as visual images, gas concentration data, environmental characteristics, and maturity level. This provides precise pest warnings and control recommendations for crop management. For example, the pest identification results can indicate the type of pest occurring on or around the crop surface, potentially identifying different pest types such as aphids, whiteflies, and fruit borers. Each pest has its own typical visual and gas release characteristics, and the identification results will be classified according to these characteristics. The crop pest identification results will provide the specific location of the pest occurrence, helping farmers to accurately know which part of the crop is affected by the pest. For example, the pest may only affect a part of the fruit or be concentrated in the leaf area. The regional information provided by the identification results allows control measures to be more precisely targeted at the affected area, rather than spraying pesticides on a large scale, thus avoiding unnecessary waste of resources.

[0074] The above process not only improves the accuracy of pest identification, but also helps farmers take timely control measures to minimize crop losses.

[0075] In this embodiment, a field experiment was conducted using citrus pest and disease identification as an example. By combining crop visual recognition technology, gas sensing technology, and environmental feature analysis, accurate identification of citrus pests was successfully achieved. First, based on real-time collected crop surface image data, the system extracted parts containing abnormal features (such as fruit borer moth holes, aphid traces, etc.) using an image segmentation module, and enhanced the salience of the pest-affected areas using an abnormal feature enhancement module. Simultaneously, based on synchronously collected volatile gas concentration data, the system analyzed the relationship between gas changes and crop maturity. Combining the concentration changes of gases such as ethylene and ethanol, the maturity and expected maturity of the citrus were determined. By linking environmental characteristics, maturity, and external pest characteristics of crops, the system accurately identified citrus trees as being infested by fruit borers, aphids, and whiteflies. It assessed the severity and distribution of each pest, finding that fruit borer damage accounted for 20% of the affected area, aphid damage for 5%, and whitefly damage for 10%. Based on the pest identification results, the system provided precise control recommendations, such as recommending the use of biological pesticides for fruit borers and strengthening the control of aphids and whiteflies. Ultimately, through dynamically optimized pest control strategies, the system improved control precision by more than 30%, effectively reducing crop losses and enhancing the timeliness and accuracy of pest control.

[0076] Example 2: A visual recognition-based crop pest identification system, see [link to example]. Figure 1 As shown, it includes:

[0077] The crop visual processing module includes a visual processing unit and an auxiliary processing unit. The visual processing unit is used to acquire crop image data in real time based on visual devices to obtain crop surface image data; construct a crop visual recognition model; and process the crop surface image data based on the crop visual recognition model to obtain crop abnormal surface images and crop growth stage characteristics.

[0078] The auxiliary processing unit is used to synchronously collect the concentration of volatile gases around crops based on gas sensing devices; to identify the crop status based on the volatile gas concentration, and to obtain the crop maturity characteristics; the crop maturity characteristics are used to represent the maturity stage and expected maturity time of the crop.

[0079] The crop pest identification module includes a pest identification unit; the pest identification unit is used to acquire crop environmental characteristics; based on crop environmental characteristics, crop maturity characteristics and abnormal crop surface images, pest identification is performed on the crop to obtain crop pest identification results.

[0080] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for identifying crop pests based on visual recognition, characterized in that, Includes the following steps: Real-time acquisition of crop image data using vision devices yields crop surface image data; a crop visual recognition model is constructed; and the crop surface image data is processed based on the crop visual recognition model to obtain abnormal crop surface images and characteristics of crop growth stages. The concentration of volatile gases around crops is collected synchronously using gas sensing devices; the crop status is identified based on the volatile gas concentration to obtain crop maturity characteristics; the crop maturity characteristics are used to represent the maturity stage and expected maturity time of crops. Obtain crop environmental characteristics; identify crop pests based on crop environmental characteristics, crop maturity characteristics, and abnormal crop surface images, and obtain crop pest identification results; The specific steps for identifying crop status based on volatile gas concentrations include: Data processing based on volatile gas concentrations yields standard time-series gas concentration data; the mean gas concentration of the standard time-series gas concentration data is compared with the corresponding volatile gas threshold range of crops to obtain the basic maturity level of crops. Meanwhile, the concentrations of the first and second gases are extracted from the standard time-series gas concentration data and compared to obtain the active gas ratio. When the active gas ratio exceeds the preset gas ratio, the corresponding crop is judged to be in the respiratory climax period; otherwise, the corresponding crop is in the respiratory plateau period. At the same time, the diurnal variation pattern of the third gas concentration is analyzed to obtain the volatile substance release characteristics. A gas concentration change trend model is established based on the respiratory climax or respiratory stationary phase. By continuously monitoring the slope of the changes in the first, second, and third gas concentrations, the characteristics of the predicted crop ripening process can be obtained. The characteristics of crop maturity are obtained by fusing the basic maturity level of crops, the characteristics of volatile substance release, and the characteristics of predicted crop maturity process. The specific steps for identifying crop pests based on crop environmental characteristics, crop maturity characteristics, and abnormal crop surface images include: Correlation analysis is performed based on crop environmental characteristics and crop maturity characteristics to obtain external characteristics of crop diseases; the identification sensitivity of different types of pests is matched according to the external characteristics of crop diseases. By analyzing factors such as temperature, humidity, light intensity, and soil moisture in the environment where crops are located, and combining these with the maturity characteristics of crops, we can understand the current growth status of crops and their relationship with the environment. The external characteristics of diseases, including the indicative changes on the crop surface, are correlated with the types of pests. By combining existing pest data, the identification sensitivity of the corresponding pests can be matched. By adjusting the identification sensitivity based on changes in the environment and external characteristics of diseases, the sensitivity of identifying specific pest types can be improved. The pest identification threshold for abnormal crop surface images is adjusted based on the identification sensitivity to obtain the corrected pest distribution. The crop environmental characteristics and the corrected pest distribution are spatially correlated and combined with the crop maturity characteristics to identify crop pests, thus obtaining the crop pest identification results. Based on the recognition sensitivity, the pest recognition threshold for abnormal surface images of crops is corrected; by dynamically adjusting the pest recognition threshold, adaptive adjustments are made to meet the pest recognition needs of different environments, different crop varieties, and different maturity stages; after the pest recognition threshold is corrected, a pest distribution map is obtained, reducing the phenomenon of false recognition and missed recognition; the corrected pest distribution map reflects the pest areas that exist in certain specific areas or growth stages of crops. Spatial correlation analysis is used to identify the distribution patterns of pests in farmland or orchards, and a comprehensive assessment is made in conjunction with the maturity characteristics of crops. By combining spatial data of environmental characteristics, modified pest distribution, and maturity characteristics, a comprehensive pest identification of crops is carried out, and the crop pest identification results are obtained. The crop pest identification results specifically indicate whether pests or diseases exist on the crop surface or in the surrounding environment, as well as the type, location, severity, and affected area of ​​the pests, by analyzing the visual images of crops, gas concentration data, environmental characteristics, and maturity. The volatile gas threshold range and preset gas ratio are set based on the physiological characteristics of crops and their natural relationship with environmental changes, rather than values ​​set arbitrarily by humans.

2. The method for identifying crop pests based on visual recognition according to claim 1, characterized in that, The specific steps for processing crop surface image data based on a crop visual recognition model include: The crop visual recognition model includes an image segmentation module, an anomaly feature enhancement module, and a growth feature judgment module; In the image segmentation module, the crop surface image data is segmented by abnormal features, and the crop abnormal image containing abnormal features and the crop background image containing a large background are extracted. In the abnormal feature enhancement module, the crop abnormal image is enhanced to obtain the crop abnormal surface image. In the growth feature judgment module, the crop growth stage is identified in the background image of the crop to be identified, and the crop growth stage features are obtained. Specifically, in the growth feature judgment module, multi-scale convolutional feature extraction is used to extract features layer by layer from the crop region of the background image of the crop to be identified, and the crop growth features are obtained. The color information in the crop growth features is extracted to obtain the crop growth color features. The trained classifier is used to judge the crop growth color features to obtain the crop growth stage features.

3. The method for identifying crop pests based on visual recognition according to claim 2, characterized in that, The specific steps for constructing an anomaly feature enhancement module in a crop visual recognition model include: The YOLOv8 model is used as the base model for the anomaly feature enhancement module in the crop visual recognition model. In the backbone network of the base model, ECA attention mechanism is introduced as a convolutional feature extraction unit, a multi-task learning branch is added to the head part as a crop feature classification unit, and CSPNet structure is used as a crop feature fusion unit in the neck part. At the same time, the Alpha-IoU function is set as the loss function to obtain the optimized anomaly feature enhancement module. The abnormal feature enhancement module processes the crop image to be identified as abnormal to obtain an abnormal surface image of the crop.

4. The method for identifying crop pests based on visual recognition according to claim 3, characterized in that, For different varieties of crops, the first gas, the second gas, and the third gas represent different types of gases.

5. A crop pest identification system based on visual recognition, characterized in that, The system employs a visual recognition-based crop pest identification method according to any one of claims 1-4, comprising: The crop visual processing module includes a visual processing unit and an auxiliary processing unit. The visual processing unit is used to acquire crop image data in real time based on visual devices to obtain crop surface image data; construct a crop visual recognition model; and process the crop surface image data based on the crop visual recognition model to obtain crop abnormal surface images and crop growth stage characteristics. The auxiliary processing unit is used to synchronously collect the concentration of volatile gases around crops based on gas sensing devices; to identify the crop status based on the volatile gas concentration, and to obtain the crop maturity characteristics; the crop maturity characteristics are used to represent the maturity stage and expected maturity time of the crop. The crop pest identification module includes a pest identification unit; the pest identification unit is used to acquire crop environmental characteristics; based on crop environmental characteristics, crop maturity characteristics and abnormal crop surface images, pest identification is performed on the crop to obtain crop pest identification results.

Images