Corn silking stage high temperature stress diagnosis method and system based on image recognition

Abstract

The application relates to the technical field of intelligent agriculture, and discloses a corn silking period high-temperature stress diagnosis method and system based on image recognition. The method synchronously acquires a visible light and thermal infrared image pair of a corn plant through integrated acquisition equipment, and performs space-time registration. The visible light image is subjected to semantic segmentation, and silk and tassel pixel regions are identified. According to the segmentation result, a first temperature distribution feature of the silk and a second temperature distribution feature of the tassel are extracted from the thermal infrared image, and meanwhile, a color change feature and a morphological feature of the silk are extracted from the visible light image. These multi-source features are combined to form a comprehensive feature set. Finally, logic judgment is performed based on the feature set and preset thresholds of various types, the high-temperature stress grade is determined, and a corresponding field water regulation strategy is matched and output. The application realizes automatic and accurate diagnosis and decision support for high-temperature stress of key reproductive organs of corn, and effectively improves the intelligence and stress resistance of field management.

Description

Technical Field

[0001] This application relates to the field of smart agriculture technology, and in particular to a method and system for diagnosing high temperature stress during the silking stage of maize based on image recognition. Background Technology

[0002] The silking stage of maize is the most sensitive growth stage to high temperature stress. High temperatures at this time will directly affect the activity of silks and the pollination and fertilization process, ultimately leading to a reduction in the number of grains per ear and a severe drop in yield. Therefore, rapid and accurate diagnosis of high temperature stress during this critical period and timely and effective water regulation are important links to ensure stable and high maize yields.

[0003] Traditional field stress diagnosis relies primarily on the experience of agricultural technicians through visual inspection and handheld single-point measurements, such as observing leaf curling or using a thermometer to measure leaf temperature. These methods have significant limitations: First, manual inspection is inefficient and highly subjective, making it difficult to conduct rapid surveys of large fields; second, as a key reproductive organ that directly determines yield, the filaments are small and hidden within the canopy, making it difficult to conduct non-destructive and accurate observations using traditional visual and contact measurements; third, single indicators such as leaf temperature or air temperature cannot comprehensively reflect the overall physiological state of crop organs, let alone quantify the direct heat damage caused by high temperatures to the filaments and the resulting secondary physiological and morphological changes.

[0004] In recent years, remote sensing and image recognition technologies have begun to be applied to crop monitoring, such as using drones equipped with thermal infrared cameras to acquire canopy temperature distribution. However, existing technologies are mostly limited to the canopy scale and cannot accurately attribute thermal signals to specific organs such as silks, resulting in insufficient specificity and accuracy in diagnosis. Furthermore, relying solely on single information sources such as temperature or visible light images lacks deep fusion of multimodal information, making it difficult to distinguish the effects of high-temperature stress from other adverse conditions (such as drought) and to assess different stages of stress development. Therefore, there is an urgent need for a non-invasive, automated technology that can focus on maize silk organs and integrate their physical temperature and visual phenotypic information to achieve accurate diagnosis and grading decision-making for high-temperature stress during the silking stage. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of this application provide a method for diagnosing high temperature stress during the silking stage of maize based on image recognition to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the image recognition-based diagnostic method for high-temperature stress during the silking stage of maize provided in this application includes:

[0007] Synchronous image pairs of corn plants in the silking stage are acquired simultaneously, the synchronous image pairs including spatiotemporally aligned visible light images and thermal infrared images;

[0008] The visible light image is subjected to image semantic segmentation processing to obtain a segmentation result image that identifies the filament pixel region and the tassel pixel region;

[0009] Based on the segmentation result image, a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region are extracted from the thermal infrared image.

[0010] Based on the segmentation result image, the color change features and morphological features of the filament pixel region are extracted from the visible light image;

[0011] The first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature are combined to form a comprehensive feature set for characterizing the high temperature stress state of a single maize plant.

[0012] Based on the calculation results of various features in the comprehensive feature set, logical judgment processing is performed on multiple preset temperature thresholds, color change thresholds, and morphological change thresholds. The high temperature stress level is determined according to the combination of satisfied threshold conditions, wherein the high temperature stress level is associated with a preset field water regulation strategy.

[0013] To address the aforementioned problems, this application also provides an image recognition-based diagnostic system for high-temperature stress during the silking stage of maize, the system comprising:

[0014] A multimodal image synchronous acquisition module is used to synchronously acquire synchronous image pairs of corn plants in the silking stage, wherein the synchronous image pairs include spatiotemporally aligned visible light images and thermal infrared images.

[0015] The semantic segmentation and organ recognition module is used to perform image semantic segmentation processing on the visible light image to obtain a segmentation result image that identifies the filament pixel region and the tassel pixel region.

[0016] The directional temperature feature extraction module is used to extract, based on the segmentation result image, a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region from the thermal infrared image;

[0017] The visual phenotypic feature extraction module is used to extract the color change features and morphological features of the filament pixel region from the visible light image based on the segmentation result image;

[0018] The multi-source feature fusion module is used to combine the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature to form a comprehensive feature set for characterizing the high temperature stress state of a single maize plant.

[0019] The stress level decision and strategy output module is used to perform logical judgment processing based on the calculation results of various features in the comprehensive feature set and multiple preset temperature thresholds, color change thresholds and morphological change thresholds, and determine the high temperature stress level according to the combination of satisfied threshold conditions. The high temperature stress level is associated with a preset field water regulation strategy.

[0020] This invention brings significant technological advancements and beneficial effects through multimodal image synchronous acquisition, organ-oriented feature extraction guided by semantic segmentation, and rule-based decision-making based on multi-source information fusion. First, this invention creatively uses spatiotemporally aligned visible light and thermal infrared image pairs as the basis for analysis, and utilizes deep learning semantic segmentation technology to accurately identify the pixel regions of filaments and tassels, achieving precise focusing from the canopy scale to the key organ scale. Based on this, the system directionally extracts the temperature distribution features of filaments and tassels from the thermal infrared images, and extracts the color and morphological change features of filaments from the visible light images, thereby constructing a comprehensive feature set integrating physical temperature, physiological response, and morphological structure information. This deep fusion of multimodal and multidimensional features overcomes the one-sidedness of diagnosis from a single information source, achieving a more comprehensive and accurate quantitative characterization of high-temperature stress.

[0021] Secondly, this invention achieves the automated and intelligent transformation from complex feature sets to clear agronomic decisions by using multi-feature thresholds calibrated in field trials and predefined logical rules. The system can automatically determine different levels of high temperature stress, from mild to severe, based on the comparison results of various indicators in the comprehensive feature set with preset thresholds, and directly associate each level with preset, specific field water control strategies. This not only greatly improves the objectivity and consistency of diagnosis and reduces reliance on human experience, but also forms a rapid closed loop of "perception-diagnosis-decision-making," transforming traditional post-event remediation into precise pre-event warning and immediate intervention, significantly improving the intelligence level and resilience of field water management. Attached Figure Description

[0022] Figure 1 A flowchart illustrating a method for diagnosing high-temperature stress during the silking stage of maize based on image recognition, provided in an embodiment of this application.

[0023] Figure 2 A functional block diagram of a corn silking stage high temperature stress diagnosis system based on image recognition provided in an embodiment of this application;

[0024] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0026] This application provides an image recognition-based method for diagnosing high-temperature stress during the silking stage of maize. The execution entity of this image recognition-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the image recognition-based method for diagnosing high-temperature stress during the silking stage of maize can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms.

[0027] Reference Figure 1 The diagram shown is a flowchart illustrating a method for diagnosing high-temperature stress during the silking stage of maize based on image recognition, according to an embodiment of this application. In this embodiment, the method includes:

[0028] S1. Synchronously acquire a pair of images of corn plants in the silking stage, the pair of images including spatiotemporally aligned visible light images and thermal infrared images.

[0029] In some embodiments, the synchronous acquisition of synchronized image pairs of maize plants in the silking stage includes spatiotemporally aligned visible light and thermal infrared images, specifically including:

[0030] The visible light image and the thermal infrared image of the corn plant are acquired simultaneously by an integrated acquisition device, wherein the integrated acquisition device includes a visible light sensor and a thermal infrared sensor that are physically fixed and have parallel optical axes, and the synchronous image pair is acquired during the midday high temperature period in the field.

[0031] The visible light image and the thermal infrared image are associated with the same timestamp and device spatial location information;

[0032] Based on pre-stored calibration parameters, the visible light image and the thermal infrared image are spatially registered to generate a pixel-aligned synchronous image pair.

[0033] In this embodiment, step S1 aims to acquire core raw data for subsequent high-temperature stress diagnostic analysis, namely, synchronized image pairs. This step solves the fundamental technical problem of strictly aligning multimodal sensing data in time and space, providing the possibility for subsequent accurate diagnosis by fusing visible light morphology information and thermal infrared temperature information.

[0034] In this embodiment, a synchronized image pair refers to a data pair consisting of a visible light image and a thermal infrared image acquired at the same time from the same spatial perspective for the same corn plant. The visible light image is image data captured by an imaging sensor capable of sensing the visible spectrum, which mainly reflects the apparent visual information such as the color, texture, and shape of the corn plant and its organs. The thermal infrared image is image data captured by an imaging sensor capable of sensing the long-wave infrared radiation emitted by an object, and the value of each pixel directly corresponds to the radiation temperature of the object's surface, thereby reflecting the surface temperature distribution of different parts of the corn plant. Spatiotemporal alignment is a technical requirement describing the relationship between this pair of images. It requires that the two images not only be acquired at exactly the same time, but also that, after geometric correction processing, the pixels representing the same physical point in the images can also correspond precisely in position.

[0035] In this embodiment, step S1 is implemented through an integrated hardware acquisition system and a preset data processing flow. First, a specially designed integrated acquisition device is used to simultaneously acquire visible light and thermal infrared images of corn plants. The key feature of this integrated acquisition device in terms of hardware construction is that its visible light sensor and thermal infrared sensor are physically fixed to the same rigid bracket or shell, and the optical axes of the two sensors are kept parallel through precise installation. This physical fixing and parallel optical axis design ensures that the two sensors have a basically consistent field of view to the greatest extent from the hardware level, which greatly simplifies the complexity of subsequent software spatial registration and is a creative hardware foundation for ensuring data spatial consistency. The acquisition is deliberately scheduled to be carried out during the midday high temperature period in the field, such as the hours when local solar radiation is strongest and the temperature is highest. This is because this stage is the critical time window when corn suffers the most significant high temperature stress during the silking stage and its physiological response is most easily detected. Targeted data acquisition at this time can effectively capture stress signals.

[0036] In this embodiment, at the moment the image is acquired, the system associates the simultaneously generated visible light image file and thermal infrared image file with the same timestamp and the same device spatial location information. The timestamp records the precise moment of image acquisition, and the device spatial location information is typically provided by an integrated GPS module, recording the geographic coordinates of the acquisition point. This operation enables data association and traceability management in both the time and macro-spatial dimensions.

[0037] In this embodiment, spatial registration of two images based on pre-stored calibration parameters is the core technology for generating the final usable synchronized image pair. The pre-stored calibration parameters are obtained through laboratory or field calibration experiments on the integrated acquisition device. These parameters include the internal parameters of the two sensors, their relative position and orientation parameters, and lens distortion coefficients. Spatial registration utilizes these calibration parameters and a computational geometric transformation model to map corresponding points in the visible light image and the thermal infrared image to the same coordinate system. For example, a perspective transformation model can be used. By calling image transformation functions from computer vision libraries such as OpenCV, the thermal infrared image can be resampled and geometrically distorted, ensuring that each pixel is aligned with its corresponding scene point in the visible light image. After this processing, the resulting pixel-aligned synchronized image pair means that any pixel of a corn organ identified in the visible light image can be directly matched with its corresponding pixel representing the temperature value of that point in the thermal infrared image, thus achieving pixel-level spatial information fusion.

[0038] In this embodiment, the technical effect of step S1 is to provide a high-quality data foundation for the entire high-temperature stress diagnosis method, which can be directly fused and analyzed. By creatively adopting a physically fixed and optically parallel dual-sensor integration scheme, and combining it with a spatial registration algorithm based on calibration parameters, this step fundamentally solves the technical problem of difficulty in accurately aligning visible light and thermal infrared images due to sensor separation and different viewing angles. This ensures that the filament and tassel regions obtained from the visible light image segmentation in subsequent steps can be mapped onto the thermal infrared image without deviation to extract accurate temperature features. This is a prerequisite for achieving accurate correlation and diagnosis of multimodal information. At the same time, selecting to collect data during the midday high-temperature period ensures that the obtained data contains the most significant stress information, improving the sensitivity and reliability of the diagnosis.

[0039] S2. Perform image semantic segmentation processing on the visible light image to obtain a segmentation result image that identifies the filament pixel region and the male spike pixel region.

[0040] In some embodiments, performing image semantic segmentation processing on the visible light image to obtain a segmentation result image that identifies the filament pixel region and the tassel pixel region specifically includes:

[0041] The visible light image is subjected to pixel-level classification processing to generate a pixel category map, wherein the pixel category map includes pixel labels for filament category, spike category and background category;

[0042] Based on the pixel category mapping, all pixels labeled as filament category are extracted to generate the filament pixel region, and all pixels labeled as male spike category are extracted to generate the male spike pixel region.

[0043] The filament pixel region and the male spike pixel region are combined to form the segmentation result image.

[0044] In this embodiment, step S2 aims to automatically and accurately identify and separate the visual regions of the two key reproductive organs most sensitive to high-temperature stress—the filament and the tassel—from the visible light image. This step is crucial for transforming the raw image data into data with clear agronomic semantic information and is the foundation for subsequent targeted extraction of temperature and morphological features.

[0045] In this application embodiment, image semantic segmentation is a technical process of classifying images at the pixel level using computer vision algorithms. The segmentation result image is a digital image with the same size as the input image, but each pixel is assigned a specific category label. In this application, these labels are specifically used to indicate whether the pixel belongs to corn silk, corn tassel, or background.

[0046] In this embodiment, step S2 relies on a pre-trained deep learning semantic segmentation model and subsequent pixel region extraction and combination logic. First, the registered visible light image from step S1 is input into a pre-trained deep learning segmentation model. This model is trained on a massive dataset of labeled cornfield images. In this dataset, each pixel of each image is manually labeled with category labels such as "stigma," "tassel," "leaf," "stem," and "soil." The model learns a complex mapping relationship from the original pixel value to the semantic category through a convolutional neural network. When a new visible light image is input, the model performs pixel-level classification processing, calculates the probability of each pixel in the output image belonging to each category, and takes the category with the highest probability as the final label of the pixel, thereby generating a pixel category mapping map. This mapping map is a two-dimensional array in memory, where each position stores an integer value representing the category.

[0047] In this embodiment, based on the generated pixel category map, the system performs a specific pixel region extraction operation. Specifically, the system traverses the entire pixel category map, indexing the coordinates of all pixels labeled as "filaments," and the set of these coordinates constitutes the filament pixel region in memory. Similarly, the coordinates of all pixels labeled as "tassels" are indexed, forming the tassel pixel region. This is a data processing process based on index lookup.

[0048] In this embodiment, the two extracted pixel regions are finally combined to form the final segmentation result image. This combination is not a simple image overlay, but rather the creation of a new, unified segmentation result image data structure at the data structure level. This data structure explicitly records all pixel coordinates covered by the filament pixel region and the tassel pixel region, and may be visualized and stored in a single-channel image with different label values ​​(e.g., using the number 1 to represent the filament and the number 2 to represent the tassel). This result image is a structured target region indicator file that can be directly called upon in subsequent steps.

[0049] In this embodiment, the technical effect of step S2 is to achieve the first intelligent parsing of unstructured image data, converting the image into a machine-understandable information map containing the precise location and shape of the target object. It creatively and accurately solves the core problem of "automatically locating key stress-response organs in complex field backgrounds." By applying a pre-trained deep learning model, this method overcomes the shortcomings of traditional image processing techniques, such as strong dependence on color and shape rules and poor robustness in complex natural scenes. It can adapt to changes in different lighting, varieties, and growth states, ensuring high-precision and high-stability identification of the filament and tassel regions, which is an indispensable prerequisite for subsequent accurate feature extraction.

[0050] S3. Based on the segmentation result image, extract the first temperature distribution feature corresponding to the filament pixel region and the second temperature distribution feature corresponding to the male spike pixel region from the thermal infrared image.

[0051] In some embodiments, extracting a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region from the thermal infrared image based on the segmentation result image specifically includes:

[0052] Based on the spatial coordinates of the filament pixel region and the tassel pixel region in the segmentation result image, the corresponding mapping region in the thermal infrared image is determined.

[0053] The temperature values ​​of all pixels are read from the corresponding mapping regions of the thermal infrared image, respectively.

[0054] Calculate the average temperature of the mapped region corresponding to the filament pixel region, and calculate the percentage of pixels in the region whose temperature value exceeds the air temperature at the same time by 5°C. Use the average temperature and the percentage together as the first temperature distribution feature.

[0055] Based on the temperature value of the mapped region corresponding to the male ear pixel region, a second statistical feature set is calculated as the second temperature distribution feature.

[0056] In this embodiment, step S3 aims to extract key temperature features from the thermal infrared image that can quantitatively characterize the thermal state of the filaments and tassels, based on the precise organ localization information provided in step S2. This step is the core link in realizing information fusion of spatially registered multimodal data. Its innovation lies in using the precise results of semantic segmentation as an index to achieve directional statistics and in-depth analysis of the temperature field of specific organs, thereby transforming thermal imaging information into feature indicators with clear agronomic diagnostic significance.

[0057] In this embodiment, step S3 is implemented as a data processing flow based on coordinate mapping and statistical calculation. First, the system directly determines the corresponding mapping regions in the registered thermal infrared image based on the spatial coordinates of the filament pixel region and the tassel pixel region recorded in the segmentation result image. Since step S1 has ensured that the pixels of the two images are aligned, this determination process is a direct and accurate coordinate indexing operation. That is, each target pixel coordinate in the segmentation result image corresponds to the pixel coordinates of the same spatial location in the thermal infrared image.

[0058] In this embodiment, the system then reads the temperature values ​​of all pixels from the two corresponding mapping regions identified in the thermal infrared image. This is a data retrieval process. For the filament mapping region, a set of N temperature values ​​is obtained, and for the tassel mapping region, another set of M temperature values ​​is obtained. These temperature values ​​are the original data basis for all subsequent feature calculations.

[0059] In this embodiment, a first temperature distribution feature is calculated for the set of temperature values ​​in the filament pixel region. This application creatively defines a feature set consisting of two complementary indices. The first index is the arithmetic mean of all temperature values ​​in the region, reflecting the overall average temperature level of the filament organ surface. The second index is the percentage of pixels in the region whose temperature exceeds the ambient air temperature by 5 degrees Celsius. The calculation process involves first obtaining the ambient air temperature at the time of image acquisition from field environmental records, then setting a threshold of 5 degrees Celsius above the ambient air temperature, then counting the number of temperature values ​​in the filament region set that exceed this threshold, and finally dividing this number by the total number of pixels N in the filament region and multiplying by 100% to obtain the percentage. This combination of two indices macroscopically characterizes the overall thermal state of the filament and microscopically quantifies the severity of local overheating, providing a more comprehensive picture of the thermal response of the filament tissue under high-temperature stress.

[0060] In this embodiment, a second temperature distribution feature is calculated for the set of temperature values ​​in the tassel pixel region. The second temperature distribution feature is a statistical feature set, the purpose of which is to describe the temperature distribution of the tassel from different statistical dimensions. The calculation of this feature set is based on the set of temperature values ​​in the tassel region, and may include, but is not limited to, calculating the mean, standard deviation, maximum value, minimum value, specific quantile, and other statistical quantities of the set. As a reference organ for corn on the same plant, the temperature distribution feature of the tassel can be used to reflect the overall condition of the canopy microclimate and provide background reference for assessing the degree of abnormality in silk temperature.

[0061] In this embodiment, the technical effect of step S3 is to achieve the directional transformation and quantification from thermal infrared images to diagnostic features. By creatively employing the method of "semantic segmentation localization and temperature field directional extraction," this step solves the technical problem of automatically and accurately separating the temperature information of specific organs in traditional thermal imaging analysis. The extracted first temperature distribution feature (the ratio of average temperature to high-temperature pixels) directly and sensitively reflects the physical intensity and spatial range of thermal stress suffered by the most vulnerable organ, the filament, and is one of the most core input variables in the diagnostic model. The second temperature distribution feature provides important contextual information for diagnosis.

[0062] S4. Based on the segmentation result image, extract the color change features and morphological features of the filament pixel region from the visible light image.

[0063] In some embodiments, extracting the color change features and morphological features of the filament pixel region from the visible light image based on the segmentation result image specifically includes:

[0064] Based on the segmentation result map, locate the filament pixel region in the visible light image;

[0065] Read the color channel values ​​of the pixels from the filament pixel region of the visible light image;

[0066] The intensity ratio of the red channel to the green channel in the RGB color space of the filament pixel region is calculated based on the color channel values, and used as the color change feature.

[0067] The proportion of the pixel area of ​​the filament pixel region to the total pixel area of ​​the maize plant canopy projection in the visible light image is calculated as the morphological feature.

[0068] In this embodiment, step S4 aims to quantify and extract the physiological and morphological response characteristics of the filamentous organs under high-temperature stress from visible light images. This step creatively utilizes the precisely located filamentous region information from the same data source (visible light images) to extract visual phenotypic parameters directly related to stress, serving as an important supplement to the temperature physical characteristics obtained in step S3, and jointly constructing a multi-dimensional diagnostic basis.

[0069] In this embodiment, step S4 is implemented by performing color space analysis and area ratio calculation on the located filament region. First, based on the segmentation result image output in step S2, the system can directly locate the filament pixel region in the visible light image. Since the segmentation result image is essentially a set of coordinate indices of the target pixels, this "location" operation directly calls this coordinate set.

[0070] In this embodiment, the system then reads the color channel values ​​of each pixel within the filament pixel region from the visible light image. Specifically, for a visible light image in RGB format, the system reads the intensity values ​​of the red, green, and blue color channels for each pixel at specified coordinates; these values ​​are typically between 0 and 255.

[0071] In this embodiment, the red-green channel intensity ratio, serving as a color change feature, is calculated based on the read color channel values. This is achieved by first calculating the average red channel intensity and the average green channel intensity of all pixels within the filament region. Then, the average red channel intensity is divided by the average green channel intensity; the quotient is the color change feature. This can be expressed as: Color Change Feature = Average Red Channel Intensity / Average Green Channel Intensity. For example, if the calculated average red channel intensity is 130 and the average green channel intensity is 80, the ratio is 1.625. The red-green ratio is chosen as the feature based on agronomical knowledge: high temperature and drought stress often lead to water loss and decreased activity in the filaments, accompanied by a brownish color transition. This process manifests in the RGB color space as a relative increase in the red component and a relative decrease in the green component, thus increasing the ratio. This indicator is more stable than a single color channel value in capturing stress-related color trend changes.

[0072] In this embodiment, the proportion of the filament region pixel area (a morphological feature) to the total canopy projection area is calculated simultaneously. This is achieved through two parts: First, the pixel area of ​​the filament region is calculated, which can be obtained by directly counting the total number of pixels belonging to the filament category in the segmentation result image. Second, the total pixel area of ​​the canopy projection of the entire maize plant in the visible light image needs to be obtained. This can be achieved by performing another round of semantic segmentation on the visible light image or using image processing methods such as color indexing and threshold segmentation to separate the maize plant canopy (including all leaves, stems, tassels, and filaments) from the soil background, thereby counting the total number of pixels in the canopy region. Finally, the morphological feature is obtained by dividing the filament pixel area by the total canopy pixel area. For example, if the filament pixel area is 1500 and the total canopy pixel area is 20000, the ratio is 0.075. Using a ratio rather than an absolute area effectively eliminates the influence of scale variations caused by different shooting distances and angles, making the feature more robust and comparable. This feature can reflect the fullness of filament development or the degree of shrinkage after stress.

[0073] In this embodiment, the technical effect of step S4 is to obtain visual quantitative indicators that can intuitively reflect the physiological state and morphological structure of filaments under high-temperature stress. The color change feature transforms the phenomenon of "darkening and browning" that is observable to the naked eye but difficult to quantify into a specific numerical value, realizing an objective assessment of the physiological state of filaments. The morphological feature transforms the degree of "wilt and shrinkage" of filaments into an area ratio parameter, realizing an objective measurement of the degree of morphological damage. These two features, together with the temperature feature extracted in step S3, characterize the impact of high-temperature stress from different dimensions (physiological, morphological), greatly enriching the information dimensions of diagnosis.

[0074] S5. The first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature are combined to form a comprehensive feature set for characterizing the high temperature stress state of a single maize plant.

[0075] In some embodiments, combining the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature to form a comprehensive feature set for characterizing the high-temperature stress state of a single maize plant specifically includes:

[0076] The system receives the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature as input data.

[0077] The input data is organized according to a predetermined fusion order and data structure to generate a multidimensional data sequence, wherein the arrangement of various features in the multidimensional data sequence is fixed.

[0078] In this embodiment, step S5 aims to systematically integrate the diagnostic indicators of different dimensions extracted independently from the multimodal images in the aforementioned steps, forming a unified, machine-processable data object, namely, a comprehensive feature set. This step is a key step in gathering scattered physical, physiological, and morphological evidence into a complete "plant stress state file." Its innovation lies in providing a standardized and information-complete input interface for subsequent rule-based logical judgments through a predetermined, structured data fusion method.

[0079] In this embodiment, step S5 is implemented through a deterministic data receiving and organization process. First, the system receives input data from steps S3 and S4. Specifically, this input data includes a first temperature distribution feature (containing the average temperature of the filaments and the proportion of high-temperature pixels) and a second temperature distribution feature (e.g., a set of statistics such as the average temperature of the tassel and the standard deviation) obtained from step S3, as well as filament color change features (the ratio of red to green channel intensity) and filament morphological features (the proportion of the filament area to the total canopy area) obtained from step S4.

[0080] In this embodiment, after receiving all input data, the system organizes them into a multidimensional data sequence according to a predetermined fusion order and data structure. The predetermined fusion order is predefined during the system design phase based on agronomic diagnostic logic and model processing requirements. For example, a reasonable order could be "filament temperature characteristics → tassel temperature characteristics → filament color characteristics → filament morphology characteristics". The arrangement of various features in this sequence is fixed, meaning that for each diagnosis, the average filament temperature always appears in the first position of the sequence, the high-temperature pixel ratio appears in the second position, and so on. The organization process is usually implemented in the program as constructing a one-dimensional array or feature vector. For example, the above features are stored sequentially in an array Feature_Integrated, the contents of which may be represented as [average filament temperature value, high-temperature pixel ratio of filament, average tassel temperature value, standard deviation of tassel temperature, red-green ratio, area ratio value]. This fixed structure ensures the consistency of the data format and is the basis for automated and batch processing.

[0081] In this embodiment, the technical effect of step S5 is to achieve the standardized fusion and encapsulation of multi-source heterogeneous features. It creatively solves the technical problem of how to effectively and unambiguously combine feature data output from different processing branches (temperature extraction, visual analysis) for final decision-making. By generating a multi-dimensional data sequence with a fixed structure, this step integrates the originally scattered temperature, color, and morphological indicators into a comprehensive feature set that can fully characterize the high-temperature stress state of a single maize plant at a specific time. This allows subsequent judgment rules or models to access and process all diagnostic information in a unified and efficient manner.

[0082] S6. Based on the calculation results of each feature in the comprehensive feature set, and the preset multiple temperature thresholds, color change thresholds and morphological change thresholds, perform logical judgment processing, and determine the high temperature stress level according to the combination of satisfied threshold conditions, wherein the high temperature stress level is associated with the preset field water regulation strategy.

[0083] In some embodiments, the calculation results based on the features in the comprehensive feature set are logically judged and processed with multiple preset temperature thresholds, color change thresholds, and morphological change thresholds. The high-temperature stress level is determined based on the combination of satisfied threshold conditions. The high-temperature stress level is associated with a preset field water regulation strategy, specifically including:

[0084] The calculation results of the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature are extracted one by one from the comprehensive feature set.

[0085] Each extracted calculation result is compared with the corresponding temperature threshold, color change threshold, or morphological change threshold to generate a set of comparison results.

[0086] Based on predefined judgment rules, each set of comparison results is analyzed. The judgment rules specify the high temperature stress level corresponding to different combinations of comparison results.

[0087] Based on the analysis results, a specific high-temperature stress level is output;

[0088] The output high temperature stress level is matched with a predefined field water regulation strategy mapping table to obtain and output the associated field water regulation strategy.

[0089] In this embodiment, step S6 aims to transform the comprehensive quantitative characteristics representing plant status constructed in the preceding steps into diagnostic conclusions and action plans with practical agronomic guidance through a set of clear decision-making rules. This step creatively realizes a closed loop from "multi-dimensional data perception" to "precision agronomic decision-making," and is the culmination of the entire method in ultimately solving the core problem of real-time diagnosis and precise control of high-temperature stress in the field.

[0090] In this embodiment, step S6 is implemented through a rule engine-based logical judgment and strategy mapping process. This process begins by extracting the specific calculation results of each feature from the comprehensive feature set generated in step S5. For example, extracting the average temperature of the filaments and the percentage of high-temperature pixels from the first temperature distribution feature, extracting the average temperature of the male spike from the second temperature distribution feature, extracting the red-green channel intensity ratio from the color change feature, and extracting the filament area ratio from the morphological feature.

[0091] In this embodiment, the result of each extracted feature calculation is then compared with a preset, corresponding threshold. These preset thresholds include multiple temperature thresholds, color change thresholds, and morphological change thresholds. For example, a base temperature threshold may be set to determine whether the filament is overheated, and a higher temperature threshold may be set to determine the degree of overheating; a color change threshold may be set to determine whether the filament shows significant browning; and a morphological change threshold may be set to determine whether the filament has significantly shrunk. The comparison operation generates a set of Boolean values ​​or comparison status results, such as "average filament temperature is greater than threshold A", "red-green ratio is greater than threshold B", "filament area ratio is less than threshold C", etc.

[0092] In this embodiment of the application, the above-mentioned combination of comparison results is parsed based on a set of predefined judgment rules. These judgment rules are explicitly specified in the form of conditional statements. Different combinations of comparison results will correspond to different levels of high-temperature stress. For example, a rule may stipulate that: if only "the average temperature of the filament is greater than threshold A" is met, it is judged as mild stress; if both "the average temperature of the filament is greater than threshold A" and "the red-green ratio is greater than threshold B" are met, it is judged as moderate stress; if all three conditions are met simultaneously, namely "the average temperature of the filament is greater than threshold A", "the red-green ratio is greater than threshold B" and "the filament area ratio is less than threshold C", it is judged as severe stress. The rule parsing process is to traverse these predefined conditions based on the currently input feature comparison results and find the rule entries that match perfectly.

[0093] In this embodiment of the application, based on the result of rule parsing, the system outputs a definite high-temperature stress level, which is the final classification conclusion of the severity of high-temperature stress suffered by the current single corn plant, such as "mild", "moderate" or "severe".

[0094] In this embodiment, the final step is to match the output high-temperature stress level with a predefined field water management strategy mapping table. This mapping table establishes the association between different stress levels and specific, actionable agronomic measures. For example, "mild stress" may be associated with "it is recommended to arrange a follow-up examination within three days," "moderate stress" with "it is recommended to implement 10 mm drip irrigation the next day," and "severe stress" with "it is recommended to implement 15 mm sprinkler irrigation immediately." The system obtains and outputs the field water management strategy corresponding to the current diagnostic level by looking up the table, thereby completing the entire chain from perception and analysis to decision output.

[0095] In this embodiment, the technical effect of step S6 is to endow the entire technical solution with final decision-making intelligence and practical value. It creatively transforms the complex multimodal feature analysis results into intuitive stress levels and specific agronomic operation instructions through a set of clear and interpretable threshold rules. This completely solves the problems of strong subjectivity and delayed response caused by the disconnect between diagnosis and action and reliance on human experience judgment in traditional methods, and realizes standardized, automated diagnosis and real-time precise intervention for high temperature stress.

[0096] In some embodiments, the step of parsing each set of comparison results based on predefined judgment rules specifically includes:

[0097] If only the average value of the first temperature distribution feature exceeds the first temperature threshold, it is determined to be mild stress.

[0098] If the average value of the first temperature distribution feature exceeds the first temperature threshold, and the color change feature exceeds the color change threshold, then it is determined to be a moderate stress level.

[0099] If the average value of the first temperature distribution feature exceeds the first temperature threshold, and the color change feature exceeds the color change threshold, and the morphological feature exceeds the morphological change threshold, then it is determined to be a severe stress level.

[0100] In this embodiment, this section provides a concretization and illustrative explanation of the key step S6, "analyzing based on predefined judgment rules." It demonstrates a creative logical criterion that hierarchically and progressively integrates multi-dimensional features to make a final diagnosis. This criterion closely simulates the gradual development process of high-temperature stress on corn silks, from the surface to the interior, from physiological response to morphological damage.

[0101] In this embodiment, the determination rule specifically refers to a predefined mapping relationship between a set of multi-feature combinations expressed in conditional logic and the high-temperature stress level. The core of this rule is that it does not rely on the simple exceeding of a single feature, but rather makes a comprehensive judgment based on whether the threshold conditions of multiple different types of features (temperature, color, morphology) are simultaneously satisfied and the combinations thereof.

[0102] In this embodiment of the application, the specific implementation of the determination rule is executed through a programmed "if-then" logic judgment process. After obtaining the comparison results of all features and thresholds, the system will perform matching according to the order and logic defined by the rule.

[0103] First, the system determines whether the condition of "the average value of the first temperature distribution feature exceeds the first temperature threshold" is met. Here, "exceeds" means that the average temperature of the filament is significantly higher than the normal or background level, indicating the physical starting point of heat stress. If other features (color, morphology) do not exceed their corresponding thresholds, it is determined to be mild stress, which corresponds to the early stage of stress, mainly manifested as increased tissue temperature, but visible physiological discoloration and morphological atrophy have not yet occurred.

[0104] Secondly, if the system determines that both the "average value of the first temperature distribution characteristic exceeds the first temperature threshold" and the "color change characteristic exceeds the color change threshold" are met simultaneously, it means that on the basis of rising temperature, the filaments have shown a significant increase in color (red-green ratio increase) due to abnormal physiological metabolism (such as dehydration and pigment changes). At this time, the system determines it to be a moderate stress level, which reflects the further development of stress, from simple physical heat damage to an observable physiological disorder stage.

[0105] Finally, if the system determines that the three conditions of "the average value of the first temperature distribution characteristic exceeds the first temperature threshold", "the color change characteristic exceeds the color change threshold", and "the morphological characteristic exceeds the morphological change threshold" are met simultaneously, it should be noted that for morphological characteristics (filament area ratio), the threshold of "exceeding" usually refers to being below the threshold of a healthy morphology in a practical physiological sense. That is, the filament has obviously wilted and shriveled due to severe water loss. When abnormalities in the three dimensions of temperature, color, and morphology occur simultaneously, it indicates that the stress is very severe and has caused comprehensive damage to the floral organs. Therefore, it is judged as a severe stress level.

[0106] In the embodiments of this application, the technical effect of this judgment rule is to achieve intelligent diagnostic logic and agronomic rationality. It creatively constructs a multi-feature, hierarchical decision tree, enabling the diagnostic conclusions to accurately correspond to different stages of stress development. Compared with methods that only use a single temperature threshold, this rule significantly improves the accuracy of diagnosis and the advance warning by introducing color and morphology as features of subsequent responses, avoiding misjudgments caused by instantaneous high temperatures in the environment. At the same time, it can also identify situations where physiological morphology has been damaged even if the instantaneous temperature is not high, but the continuous drought has caused damage. It transforms complex agronomic experience into automatically executable, objective computer logic.

[0107] In some embodiments, the preset multiple temperature thresholds, color change thresholds, and morphological change thresholds are determined in the following manner:

[0108] During the silking stage of maize, a field experiment was conducted with gradient stress treatments including normal irrigation, mild drought, moderate high temperature, and severe high temperature and drought.

[0109] During each stress treatment, multiple sets of synchronized image pairs were acquired simultaneously, and the corresponding field microenvironment air temperature and relative humidity were recorded.

[0110] After the maize matures, the final number of grains per ear for each treatment and the corresponding image acquisition unit are measured.

[0111] Based on the percentage decrease in the final number of grains per ear relative to the control group under normal irrigation, a reference stress level was assigned to each experimental plant.

[0112] For each reference stress level, the calculation results of the first temperature distribution characteristics, the second temperature distribution characteristics, the color change characteristics, and the morphological characteristics of all experimental plants under that level are statistically analyzed.

[0113] By statistically analyzing the characteristic calculation results between different reference stress levels, the temperature threshold, color change threshold, and morphological change threshold that can distinguish adjacent stress levels are determined.

[0114] In this embodiment of the application, this section reveals how the threshold system supporting the scientific and reliable operation of the entire diagnostic method is established. It elaborates in detail that the preset multiple temperature thresholds, color change thresholds, and morphological change thresholds are not subjectively set, but objectively determined through a rigorous field trial and data analysis process with the final yield loss as the reference standard. This process is the creative foundation for ensuring the accuracy and practicality of the entire diagnostic method.

[0115] In this embodiment, the threshold determination method begins with a carefully designed gradient stress field trial. This trial was conducted during the silking stage, when maize is most sensitive to high temperatures. By controlling the coupling of irrigation and natural high temperatures, multiple gradient stress levels were set up, including a normal irrigation control group, a mild drought treatment, a moderate high temperature treatment, and a severe high temperature and drought treatment. This gradient design was intended to simulate and cover various stress states that may occur in the field, ranging from normal to severely damaged, creating conditions for obtaining characteristic data at different levels.

[0116] In this embodiment of the application, during the critical period when each preset stress treatment takes effect, the method described in step S1 of this application is strictly followed to simultaneously collect multiple sets of synchronous image pairs of maize plants in the silking stage; at the same time, equipment such as a mini weather station is used to record the field microenvironment air temperature and relative humidity around the plant at the time of collection. This step ensures that all image features extracted subsequently have clear field environmental background information, and that the image data and the final yield data can be accurately associated with the same individual.

[0117] In this embodiment of the application, after the corn matures, each individual plant that participated in the image acquisition is evaluated and its final number of grains per ear is measured. The number of grains per ear is one of the most critical and stable agronomic traits for measuring the direct impact of silking stress on corn yield. Using the average number of grains per ear of all individual plants in the normal irrigation treatment control group as a benchmark, the percentage decrease in the number of grains per ear of individual plants under other treatments relative to this benchmark is calculated.

[0118] In this embodiment of the application, a reference stress level is assigned to each experimental plant based on the calculated percentage decrease in the number of grains per ear. For example, a percentage decrease less than a certain threshold can be defined as "normal / no stress", between two values ​​as "mild stress", higher as "moderate stress", and exceeding the highest threshold as "severe stress". This level, based on the final yield loss, is considered the "true value" or "reference standard" of the stress level experienced by the plant, and is used to guide subsequent characteristic analysis.

[0119] In this embodiment, for each reference stress level defined by yield loss, all test plants belonging to that level are collected. Then, the characteristic results of these plants previously calculated through steps S2 to S5 of this application are statistically analyzed. Specifically, the numerical distribution of the first temperature distribution characteristics (such as average filament temperature, high-temperature pixel ratio), second temperature distribution characteristics, color change characteristics (red-green ratio), and morphological characteristics (filament area ratio) of all plants are statistically analyzed under each reference level such as "mild", "moderate", and "severe". For example, the mean, standard deviation, and distribution range of each characteristic under each level are calculated.

[0120] In this embodiment, the final threshold for diagnosis is determined by conducting in-depth statistical analysis of the characteristic calculation results between different reference stress levels. For example, the distribution difference of the average temperature of the filaments between the "no stress" and "mild stress" groups is analyzed. Through statistical tests (such as t-tests) and effect size analysis, a temperature value that can most significantly distinguish between the two groups is found and set as the "first temperature threshold". Similarly, the difference in color change characteristics between the "mild stress" and "moderate stress" groups is analyzed to determine the "color change threshold"; the difference in morphological characteristics between the "moderate stress" and "severe stress" groups is analyzed to determine the "morphological change threshold". The statistical analysis method ensures that the selected threshold has the greatest class discrimination ability.

[0121] In this embodiment, the technical effect of this threshold determination method is to provide a solid and reliable decision-making benchmark for the entire diagnostic method. It creatively uses the final yield loss to inversely calibrate the "stress level," which is difficult to observe directly. Based on this, it mines image feature thresholds from a large amount of experimental data that can provide early warnings and are highly correlated with yield loss. This completely solves the core problem of threshold setting in rule-based systems, which relies on experience and lacks objective evidence. It ensures that every step of the diagnostic method, from feature extraction to logical judgment, is based on verifiable experimental data, significantly improving the scientific rigor and credibility of the diagnostic results.

[0122] like Figure 2 The diagram shown is a functional block diagram of a corn silking stage high temperature stress diagnosis system based on image recognition provided in an embodiment of this application.

[0123] The image recognition-based high-temperature stress diagnosis system 100 for maize during the silking stage described in this application can be installed in an electronic device. Depending on the functions implemented, the image recognition-based high-temperature stress diagnosis system 100 for maize during the silking stage may include a multimodal image synchronous acquisition module 101, a semantic segmentation and organ recognition module 102, a directional temperature feature extraction module 103, a visual phenotypic feature extraction module 104, a multi-source feature fusion module 105, and a stress level decision and strategy output module 106. The module described in this application can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0124] In this embodiment, the functions of each module / unit are as follows:

[0125] The multimodal image synchronous acquisition module 101 is used to synchronously acquire synchronous image pairs of corn plants in the silking stage, the synchronous image pairs including spatiotemporally aligned visible light images and thermal infrared images.

[0126] The semantic segmentation and organ recognition module 102 is used to perform image semantic segmentation processing on the visible light image to obtain a segmentation result image that identifies the filament pixel region and the male spike pixel region.

[0127] The directional temperature feature extraction module 103 is used to extract, based on the segmentation result image, a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region from the thermal infrared image;

[0128] The visual phenotypic feature extraction module 104 is used to extract the color change features and morphological features of the filament pixel region from the visible light image based on the segmentation result image;

[0129] The multi-source feature fusion module 105 is used to combine the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature to form a comprehensive feature set for characterizing the high temperature stress state of a single maize plant.

[0130] The stress level decision and strategy output module 106 is used to perform logical judgment processing based on the calculation results of various features in the comprehensive feature set and multiple preset temperature thresholds, color change thresholds and morphological change thresholds, and determine the high temperature stress level according to the combination of satisfied threshold conditions, wherein the high temperature stress level is associated with a preset field water regulation strategy.

[0131] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0132] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0133] Furthermore, the functional modules in the various embodiments of this 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 integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0134] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application.

[0135] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A method for diagnosing high-temperature stress during the silking stage of maize based on image recognition, characterized in that, The method includes: Synchronous image pairs of corn plants in the silking stage are acquired simultaneously, the synchronous image pairs including spatiotemporally aligned visible light images and thermal infrared images; The visible light image is subjected to image semantic segmentation processing to obtain a segmentation result image that identifies the filament pixel region and the tassel pixel region; Based on the segmentation result image, a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region are extracted from the thermal infrared image. Based on the segmentation result image, the color change features and morphological features of the filament pixel region are extracted from the visible light image; The first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature are combined to form a comprehensive feature set for characterizing the high temperature stress state of a single maize plant. Based on the calculation results of various features in the comprehensive feature set, logical judgment processing is performed on multiple preset temperature thresholds, color change thresholds, and morphological change thresholds. The high temperature stress level is determined according to the combination of satisfied threshold conditions, wherein the high temperature stress level is associated with a preset field water regulation strategy.

2. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The synchronous acquisition of images of maize plants in the silking stage includes a pair of synchronously acquired images, comprising spatiotemporally aligned visible light and thermal infrared images, specifically including: The visible light image and the thermal infrared image of the corn plant are acquired simultaneously by an integrated acquisition device, wherein the integrated acquisition device includes a visible light sensor and a thermal infrared sensor that are physically fixed and have parallel optical axes, and the synchronous image pair is acquired during the midday high temperature period in the field. The visible light image and the thermal infrared image are associated with the same timestamp and device spatial location information; Based on pre-stored calibration parameters, the visible light image and the thermal infrared image are spatially registered to generate a pixel-aligned synchronous image pair.

3. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The step of performing image semantic segmentation processing on the visible light image to obtain a segmentation result image that identifies the filament pixel region and the tassel pixel region specifically includes: The visible light image is subjected to pixel-level classification processing to generate a pixel category map, wherein the pixel category map includes pixel labels for filament category, spike category and background category; Based on the pixel category mapping, all pixels labeled as filament category are extracted to generate the filament pixel region, and all pixels labeled as male spike category are extracted to generate the male spike pixel region. The filament pixel region and the male spike pixel region are combined to form the segmentation result image.

4. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The step of extracting a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region from the thermal infrared image based on the segmentation result image specifically includes: Based on the spatial coordinates of the filament pixel region and the tassel pixel region in the segmentation result image, the corresponding mapping region in the thermal infrared image is determined. The temperature values ​​of all pixels are read from the corresponding mapping regions of the thermal infrared image, respectively. Calculate the average temperature of the mapped region corresponding to the filament pixel region, and calculate the percentage of pixels in the region whose temperature value exceeds the air temperature at the same time by 5°C. Use the average temperature and the percentage together as the first temperature distribution feature. Based on the temperature value of the mapped region corresponding to the male ear pixel region, a second statistical feature set is calculated as the second temperature distribution feature.

5. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The step of extracting the color change features and morphological features of the filament pixel region from the visible light image based on the segmentation result image specifically includes: Based on the segmentation result map, locate the filament pixel region in the visible light image; Read the color channel values ​​of the pixels from the filament pixel region of the visible light image; The intensity ratio of the red channel to the green channel in the RGB color space of the filament pixel region is calculated based on the color channel values, and used as the color change feature. The proportion of the pixel area of ​​the filament pixel region to the total pixel area of ​​the maize plant canopy projection in the visible light image is calculated as the morphological feature.

6. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The combination of the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature to form a comprehensive feature set for characterizing the high-temperature stress state of a single maize plant specifically includes: The system receives the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature as input data. The input data is organized according to a predetermined fusion order and data structure to generate a multidimensional data sequence, wherein the arrangement of various features in the multidimensional data sequence is fixed.

7. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The calculation results based on the features in the comprehensive feature set are logically judged and processed with multiple preset temperature thresholds, color change thresholds, and morphological change thresholds. The high-temperature stress level is determined according to the combination of satisfied threshold conditions. The high-temperature stress level is associated with a preset field water regulation strategy, specifically including: The calculation results of the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature are extracted one by one from the comprehensive feature set. Each extracted calculation result is compared with the corresponding temperature threshold, color change threshold, or morphological change threshold to generate a set of comparison results. Based on predefined judgment rules, each set of comparison results is analyzed. The judgment rules specify the high temperature stress level corresponding to different combinations of comparison results. Based on the analysis results, a specific high-temperature stress level is output; The output high temperature stress level is matched with a predefined field water regulation strategy mapping table to obtain and output the associated field water regulation strategy.

8. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 7, characterized in that, The predefined judgment rules are used to parse each set of comparison results, specifically including: If only the average value of the first temperature distribution feature exceeds the first temperature threshold, it is determined to be mild stress. If the average value of the first temperature distribution feature exceeds the first temperature threshold, and the color change feature exceeds the color change threshold, then it is determined to be a moderate stress level. If the average value of the first temperature distribution feature exceeds the first temperature threshold, and the color change feature exceeds the color change threshold, and the morphological feature exceeds the morphological change threshold, then it is determined to be a severe stress level.

9. The method for diagnosing high-temperature stress during the silking stage of maize based on image recognition as described in claim 1, characterized in that, The preset multiple temperature thresholds, color change thresholds, and morphological change thresholds are determined in the following way: During the silking stage of maize, a field experiment was conducted with gradient stress treatments including normal irrigation, mild drought, moderate high temperature, and severe high temperature and drought. During each stress treatment, multiple sets of synchronized image pairs were acquired simultaneously, and the corresponding field microenvironment air temperature and relative humidity were recorded. After the maize matures, the final number of grains per ear for each treatment and the corresponding image acquisition unit are measured. Based on the percentage decrease in the final number of grains per ear relative to the control group under normal irrigation, a reference stress level was assigned to each experimental plant. For each reference stress level, the calculation results of the first temperature distribution characteristics, the second temperature distribution characteristics, the color change characteristics, and the morphological characteristics of all experimental plants under that level are statistically analyzed. By statistically analyzing the characteristic calculation results between different reference stress levels, the temperature threshold, color change threshold, and morphological change threshold that can distinguish adjacent stress levels are determined.

10. A maize silking stage high-temperature stress diagnosis system based on image recognition, used to implement the maize silking stage high-temperature stress diagnosis method based on image recognition as described in any one of claims 1-9, characterized in that, The system includes: A multimodal image synchronous acquisition module is used to synchronously acquire synchronous image pairs of corn plants in the silking stage, wherein the synchronous image pairs include spatiotemporally aligned visible light images and thermal infrared images. The semantic segmentation and organ recognition module is used to perform image semantic segmentation processing on the visible light image to obtain a segmentation result image that identifies the filament pixel region and the tassel pixel region. The directional temperature feature extraction module is used to extract, based on the segmentation result image, a first temperature distribution feature corresponding to the filament pixel region and a second temperature distribution feature corresponding to the male spike pixel region from the thermal infrared image; The visual phenotypic feature extraction module is used to extract the color change features and morphological features of the filament pixel region from the visible light image based on the segmentation result image; The multi-source feature fusion module is used to combine the first temperature distribution feature, the second temperature distribution feature, the color change feature, and the morphological feature to form a comprehensive feature set for characterizing the high temperature stress state of a single maize plant. The stress level decision and strategy output module is used to perform logical judgment processing based on the calculation results of various features in the comprehensive feature set and multiple preset temperature thresholds, color change thresholds and morphological change thresholds, and determine the high temperature stress level according to the combination of satisfied threshold conditions. The high temperature stress level is associated with a preset field water regulation strategy.

Images