A method and device for monitoring, identifying and treating drought based on corn rolled leaves

Abstract

The application discloses a method and device for monitoring, identifying and processing drought based on corn leaf rolling, which can include: using a rotatable spherical camera with 4G signal transmission function to collect corn leaf images of a detection area at a first time point and a second time point, the first time point ranging from 12:00 to 14:00 in a day, and the second time point ranging from 17:00 to 19:00 in a day; using a trained corn leaf rolling detection model to determine whether the corn leaves in the corn leaf images at the first time point and the second time point are rolled, respectively, to obtain a determination result; and determining whether the corn is subjected to drought stress according to the determination result. The application is an intelligent early warning method for monitoring and identifying drought based on corn leaf rolling, which can help field managers to discover drought and flood in time and adjust irrigation time and amount.

Description

Technical Field

[0001] This invention relates to the field of crop drought identification technology, and more specifically, to a method and apparatus for identifying and addressing drought conditions based on corn leaf curl monitoring. Background Technology

[0002] Drought is one of the main factors affecting maize yield, and rapid and timely monitoring of maize drought plays a crucial role in ensuring maize production. While some existing drought monitoring methods, such as soil moisture detection, agricultural weather forecasting, and human diagnosis, can assess maize drought conditions, they are often delayed. Other methods, such as deploying immovable cameras above the monitoring area, suffer from low image clarity and unsatisfactory observation results.

[0003] In order to accurately identify corn drought, thereby providing timely and accurate information for the prevention and mitigation of corn drought stress, and improving the accuracy and timeliness of irrigation to ensure a bumper and stable corn yield, it is necessary to provide a new method for detecting whether corn is experiencing drought. Summary of the Invention

[0004] The present invention provides a method and apparatus for identifying and addressing drought conditions based on corn leaf curling monitoring, in order to overcome at least one technical problem existing in the prior art.

[0005] According to a first aspect of the present invention, a method for identifying and addressing drought conditions based on corn leaf curling monitoring is provided, comprising:

[0006] A rotatable spherical camera with 4G signal transmission capability was used to collect images of corn leaves in the area to be detected at a first time point and a second time point. The first time point was from 12 noon to 2 pm during the day, and the second time point was from 5 pm to 7 pm during the day.

[0007] The trained corn leaf curling detection model is used to determine whether the corn leaves in the corn leaf images at the first time point and the second time point are curled, and the judgment results are obtained.

[0008] If the determination result indicates that the corn leaves in the corn leaf image at the first time point are not curled, then the corn in the area to be detected is not under drought stress. If the determination result indicates that the corn leaves in the corn leaf image at the first time point are curled, and the corn leaves in the corn leaf image at the second time point are also curled, then the soil moisture sensor values ​​arranged in the soil of the area to be detected are checked. If the values ​​indicate that the soil moisture content in the area to be detected is insufficient, then the corn in the area to be detected is under a first degree of drought stress, and the soil in the area to be detected is irrigated. If the determination result indicates that the corn leaves in the corn leaf image at the first time point are curled, and the corn leaves in the corn leaf image at the second time point are not curled, then the corn in the area to be detected is under a second degree of drought stress; the first degree is greater than the second degree.

[0009] The training process of the completed corn leaf curling detection model specifically includes:

[0010] The Efficientdet object detection model is improved and used as a corn leaf curl detection model. The improvements include:

[0011] In the feature extraction process, depthwise separable convolution and inverted residual modules are used successively to improve the feature extraction capability of the backbone network in the feature extraction part;

[0012] To improve network performance, a cross-level data flow is added to the original BiFPN module of the Efficientdet object detection model.

[0013] During training, a group normalization method is used to normalize the data at the channel dimension level. After dividing the channels into several groups, the mean and variance within each group are calculated for normalization, thereby preventing gradient vanishing during model training and making it difficult for the model to converge.

[0014] A number of corn leaf images at the first and second time points are collected in advance; the corn leaf images are divided into three parts—training set, test set, and validation set—according to a predetermined ratio.

[0015] The Pseudo-Labelling method was used to iteratively train the corn leaf curling model to obtain a trained corn leaf curling detection model.

[0016] Preferably, the first time point is 2 PM in a day, and the second time point is 7 PM in a day.

[0017] Preferably, the corn leaf images are divided into three parts in a ratio of 7:2:1: training set, test set, and validation set.

[0018] Preferably, data augmentation is performed using methods such as rotation, scaling, affine transformation, and noise addition to enhance the generalization ability of the trained corn leaf curling detection model.

[0019] Preferably, the irrigation of the soil in the area to be tested specifically includes:

[0020] The system remotely controls the control switch installed on the sprinkler valve to open the valve and irrigate the soil in the area to be tested; when the soil moisture content reaches the field capacity, the system receives a corresponding alarm notification and closes the valve through the system remote control switch.

[0021] Preferably, the cross-level data flow means that, for any input node in the first layer input node of the original BiFPN module, a receiving node that receives the input information of the input node is determined in the second layer input node of the original BiFPN module, and then the previous receiving node adjacent to the receiving node is determined in the second layer input node, and a cross-level data flow is added between the input node and the previous receiving node.

[0022] According to a second aspect of the present invention, an apparatus for identifying and addressing drought conditions based on corn leaf curling monitoring is also provided, the apparatus comprising:

[0023] Metal support frame;

[0024] Several rollers are provided below the metal support frame; a rotatable spherical camera with 4G signal transmission function is provided on the crossbeam of the metal support frame, and the trained corn leaf curling detection model as described in claim 1 is deployed in the spherical camera.

[0025] The metal support frame is equipped with a cable that is connected to a 220V power supply. The cable then passes through the power cable channel of the sprinkler machine and extends along the main water conveyance beam of the sprinkler to the camera, where it is connected to the camera.

[0026] This invention provides at least the following beneficial effects from one embodiment: It is an intelligent early warning method for drought conditions based on corn leaf curling monitoring, which can help field managers promptly detect drought and flooding situations and adjust irrigation time and dosage. Simultaneously, the technical solution of this invention provides a comprehensive, blind-spot-free video monitoring method for farmland equipped with hourly sprinkler irrigation equipment, allowing cameras to observe from different locations and directions throughout the farmland. The entire farmland's status can be remotely and comprehensively monitored without physically being on-site, including real-time monitoring of the sprinkler equipment's operation and vegetation irrigation. Furthermore, a large-scale corn crop leaf curling detection model is installed on the camera; as the camera generates images, the model algorithm also detects corn leaf curling, which, combined with soil moisture content detected by a soil moisture sensor (also known as a soil humidity sensor or soil moisture sensor), helps determine whether drought has occurred. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating a method for identifying and addressing drought conditions based on corn leaf curl monitoring, provided as an embodiment of this specification;

[0029] Figure 2 This is a schematic diagram of a possible device structure in a method for monitoring, identifying, and addressing drought conditions based on corn leaf curling, as provided in the embodiments of this specification.

[0030] Figure 3 This is a schematic diagram of another possible device structure in a method for identifying and addressing drought conditions based on monitoring corn leaf curl, as provided in the embodiments of this specification.

[0031] Figure 4 for Figure 2 and Figure 3 A schematic diagram of the debugging procedure for the device shown;

[0032] Figure 5 This is a flowchart illustrating a practical scenario of a method for monitoring, identifying, and addressing drought conditions based on corn leaf curling, as provided in the embodiments of this specification.

[0033] Figure 6 A comparative schematic diagram of the principle of ordinary convolution kernel depth separable convolution in a method for monitoring, identifying and processing drought conditions based on corn leaf curling provided in the embodiments of this specification;

[0034] Figure 7 A comparative schematic diagram of the structure of the inverted residual module and the ordinary residual module used in a method for monitoring, identifying and processing drought conditions based on corn leaf curling provided in the embodiments of this specification;

[0035] Figure 8 A comparative diagram of the improved BiFPN module and the original BiFPN module of the Efficientdet target detection model in a method for monitoring, identifying and handling drought conditions based on corn leaf curling provided in the embodiments of this specification.

[0036] Figure 9 A schematic diagram of the model structure of the improved Efficientdet target detection model in a method for monitoring, identifying and processing drought conditions based on corn leaf curling provided in the embodiments of this specification;

[0037] Figure 10 This is a flowchart illustrating a method for monitoring, identifying, and addressing drought conditions based on corn leaf curling in a real-world scenario, as provided in the embodiments of this specification. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] It should be noted that the terms "comprising" and "having," and any variations thereof, in the embodiments and drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0040] As mentioned earlier, drought is one of the main factors affecting maize yield, and rapid and timely monitoring of maize drought plays a crucial role in ensuring maize production. Existing drought monitoring methods, such as soil moisture detection, agricultural weather forecasting, and human diagnosis, while capable of assessing maize drought, suffer from time lag. Other methods, such as deploying immovable cameras above the monitoring area, suffer from low image clarity and unsatisfactory observation results. In these methods, the camera is mounted on a pole or fixed at a high, immovable location. Cylindrical cameras can only observe crops within a fixed range at a fixed height and angle, resulting in a small observation range and significant limitations. While dome cameras can only rotate around a fixed position, theoretically allowing for a 360° observation range, their limited zoom capability leads to decreased image clarity for crops far from the fixed point, resulting in unsatisfactory observation results. Manual identification cannot guarantee real-time performance, cannot achieve 24-hour continuous monitoring, wastes human resources, and increases production costs. There are no standard threshold indicators for manual identification of drought conditions. It can only rely on the experience of staff. Although the appearance of large-scale leaf curling can be considered a drought, "large-scale" varies from person to person. Therefore, the indicators for each irrigation will also vary. In addition, manual identification cannot determine how much water to irrigate or when to turn off the water pump.

[0041] In order to accurately identify corn drought, thereby providing timely and accurate information for the prevention and mitigation of corn drought stress, and improving the accuracy and timeliness of irrigation to ensure a bumper and stable corn yield, it is necessary to provide a new method for detecting whether corn is experiencing drought.

[0042] The biological principles underlying this invention will be explained below. The basic principle of judging drought conditions based on corn leaf curling is that water absorbed by the corn roots reaches the leaf surface through transpiration and is then expelled through stomata. If corn encounters a certain degree of drought stress, it will curl its leaves to reduce water evaporation. Under normal circumstances, corn will curl its leaves at midday when temperatures are high and sunlight is strong. Once the sunlight weakens, the leaves will unfurl again, indicating that the corn is under drought stress, but the stress is not severe and it can recover. If the corn leaves curl up at midday and remain unfurled or incompletely unfurled by sunset, it indicates that the corn is under severe drought stress and irrigation is necessary. Leaf curling as a characteristic of corn drought is traceable.

[0043] The purpose of this invention is to provide a comprehensive, blind-spot-free video monitoring method for farmland equipped with hourly-type sprinkler irrigation equipment, allowing cameras to observe the entire farmland from different locations and directions. This enables remote and comprehensive monitoring of the entire farmland's status without physically being present, including real-time monitoring of the sprinkler equipment's operation and vegetation irrigation. Furthermore, a large-scale corn crop leaf curling detection model is installed on the camera. While the camera generates images, the model algorithm also detects corn leaf curling, and this, combined with soil moisture sensors (also known as soil humidity sensors), determines whether drought conditions are present. In the event of drought, the remote irrigation system will receive an alarm notification and remotely control the valves via a switch installed on the sprinkler valves. When the soil moisture content reaches field capacity, the system will also receive a corresponding alarm notification and remotely control the valves to close.

[0044] Next, we will provide a detailed description of a method for identifying drought conditions based on monitoring corn leaf curling, as provided in the embodiments of the specification, in conjunction with the accompanying drawings.

[0045] Figure 1 A flowchart illustrating a method for identifying drought conditions based on monitoring corn leaf curl, provided as an embodiment of this specification.

[0046] like Figure 1 As shown, the process may include the following steps.

[0047] Step 102: Use a rotatable spherical camera with 4G signal transmission capability to acquire images of corn leaves in the area to be detected at a first time point and a second time point. The first time point is from 12 noon to 2 pm during the day, and the second time point is from 5 pm to 7 pm during the day.

[0048] Step 104: Use the trained corn leaf curling detection model to determine whether the corn leaves in the corn leaf images at the first time point and the second time point are curled, and obtain the judgment results.

[0049] Step 106: If the judgment result indicates that the corn leaves in the corn leaf image at the first time point are not curled, then the corn in the area to be detected is not under drought stress; if the judgment result indicates that the corn leaves in the corn leaf image at the first time point are curled, and the corn leaves in the corn leaf image at the second time point are also curled, then the soil moisture sensor values ​​arranged in the soil of the area to be detected are checked. If the values ​​indicate that the soil moisture content in the area to be detected is insufficient, then the corn in the area to be detected is under a first degree of drought stress, and the soil in the area to be detected is irrigated; if the judgment result indicates that the corn leaves in the corn leaf image at the first time point are curled, and the corn leaves in the corn leaf image at the second time point are not curled, then the corn in the area to be detected is under a second degree of drought stress; the first degree is greater than the second degree.

[0050] Step 108: The training process of the completed corn leaf curling detection model specifically includes:

[0051] The Efficientdet object detection model is improved and used as a corn leaf curl detection model. The improvements include:

[0052] In the feature extraction process, depthwise separable convolution and inverted residual modules are used successively to improve the feature extraction capability of the backbone network in the feature extraction part;

[0053] To improve network performance, a cross-level data flow is added to the original BiFPN module of the Efficientdet object detection model.

[0054] During training, a group normalization method is used to normalize the data at the channel dimension level. After dividing the channels into several groups, the mean and variance within each group are calculated for normalization, thereby preventing gradient vanishing during model training and making it difficult for the model to converge.

[0055] Step 110: Pre-collect a number of corn leaf images at the first time point and the second time point; divide the corn leaf images into three parts: training set, test set and validation set according to a predetermined ratio;

[0056] The Pseudo-Labelling method was used to iteratively train the corn leaf curling model to obtain a trained corn leaf curling detection model.

[0057] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps may be adjusted according to actual needs, or some steps may be omitted.

[0058] based on Figure 1 In addition to the method described herein, this specification also provides some specific implementation schemes of the method, which will be described below.

[0059] In an optional embodiment, the first time point is 2 PM in a day, and the second time point is 7 PM in the same day.

[0060] In an optional embodiment, the corn leaf images are divided into three parts: a training set, a test set, and a validation set in a ratio of 7:2:1.

[0061] In optional embodiments, data augmentation is performed using methods such as rotation, scaling, affine transformation, and noise addition to enhance the generalization ability of the trained corn leaf curling detection model.

[0062] In an optional embodiment, irrigating the soil in the area to be tested specifically includes:

[0063] The system remotely controls the control switch installed on the sprinkler valve to open the valve and irrigate the soil in the area to be tested; when the soil moisture content reaches the field capacity, the system receives a corresponding alarm notification and closes the valve through the system remote control switch.

[0064] The technical solution of the present invention will be fully described below with reference to a complete embodiment.

[0065] like Figure 2 and Figure 3 As shown, Figure 2 and Figure 3 This is a schematic diagram of two variations of the device used in the method for monitoring, identifying, and handling drought conditions based on corn leaf curling provided by the present invention. The device may include a metal support frame; several rollers are arranged below the metal support frame; a rotatable spherical camera with 4G signal transmission function is arranged on the crossbeam of the metal support frame, and a trained corn leaf curling detection model is deployed in the spherical camera; a cable is arranged on the metal support frame, the cable is connected to a 220V power supply, then passes through the power cable channel of the sprinkler irrigation machine, extends along the main water conveyance crossbeam of the sprinkler irrigation machine to the camera, and is connected to the camera.

[0066] The detailed assembly method for this device is as follows:

[0067] (1) Preparation of materials and equipment:

[0068] ① Camera preparation: Select a rotatable dome camera (360° horizontally and 180° vertically) with 4G signal transmission capability as the camera to be installed.

[0069] ② Cable preparation: Copper cable with a cross-sectional area of ​​1.5 square millimeters or more, the length of which depends on the installation location of the camera.

[0070] ③ Camera mounting point preparation: Select a firm and high position as the mounting point for the dome camera based on the specific situation of the clock sprinkler equipment. If there are suitable mounting screw holes or mounting points for the dome camera, install the dome camera directly. If there are no suitable mounting screw holes or mounting points for the dome camera, welding and drilling are required to provide a suitable mounting point for the dome camera.

[0071] (2) Equipment installation:

[0072] ① Installation of the spherical camera: Install the spherical camera at the installation point at the higher position of the main water supply beam of the clockwise sprinkler system using screws or welding. The spherical camera should face downwards, ensuring that the cable can be connected to the camera along the main water supply beam of the sprinkler system.

[0073] ② Cable laying and connection: Connect the cable to the 220V power supply, then run it through the power cable channel of the sprinkler machine, and extend it along the main water conveyance beam of the sprinkler to the camera, and connect it to the camera.

[0074] (3) System debugging:

[0075] ① Camera debugging: The camera needs to be tested before and after installation. The main tests include whether the camera rotates normally, whether the switch works properly, and whether the signal is normal. The camera here uses 4G signal transmission for video transmission.

[0076] ② Video transmission signal debugging: After the camera is installed and debugged, start the video recorder to test whether the video signal is normal. If the camera transmits through the cloud, open the client (mobile phone or computer) to check whether the video signal access is normal.

[0077] ③ Overall debugging: First step, turn on the camera; second step, turn on the video recorder or video display client; third step, open the clockwise sprinkler valve; fourth step, turn on the water pump; fifth step, turn on the clockwise sprinkler rotation mode; sixth step, as the sprinkler rotates, remotely operate the camera to rotate on the video display terminal.

[0078] The preceding text described two variations of the equipment used in the method for monitoring, identifying, and addressing drought conditions based on corn leaf curling in the technical solution of this invention. The technical solution of the corn leaf curling detection model will be explained below.

[0079] The corn leaf curl detection method consists of three parts. The first part is data acquisition: cameras are installed at high points around the cornfield, and images are taken at 2 PM and 7 PM daily, capturing a balanced number of corn varieties under different environmental conditions. The second part is the corn leaf curl model: a large-scale corn crop area leaf curl detection model is implemented using a Detector without Training Parameters (DWPT) neural network detector based on the EfficientDet object detector. The third part is the corn leaf curl detection: the corn leaf curl model detects data from the cameras at different times to determine the current corn condition and thus the degree of drought.

[0080] The implementation steps are as follows:

[0081] Collect data, including images of corn leaves curling at 2 PM and 7 PM daily, showing different varieties, quantities, and different natural environmental conditions.

[0082] ② Divide the dataset into training, testing, and validation sets, using a 7:2:1 ratio;

[0083] ③ Label the data: Label the training dataset and perform data augmentation using methods such as rotation, scaling, affine transformation, and adding noise;

[0084] ④ The model is trained using an improved Efficientdet model;

[0085] ⑤ Test the model using test set data;

[0086] ⑥ The Pseudo-Labelling method was used to iteratively train the model to obtain the final corn leaf curling detection model.

[0087] The improved Efficientdet model makes the following improvements to the original model structure:

[0088] 1) Use group normalization

[0089] Batch Normalization (BN) aims to address two main issues: First, during deep neural network training, each batch has a different distribution, increasing the difficulty of model training. Second, there's the internal covariate shift (ICS) problem: during training, the activation function alters the distribution of data across layers. As the network deepens, this difference increases, leading to gradient vanishing and making model convergence difficult. Experiments using BN added to the backbone and detection subnetworks of ScratchDet for de novo training demonstrate that BN significantly mitigates gradient fluctuations during optimization, ensuring a larger learning rate and faster convergence. However, BN's limitations are also apparent. To obtain a more universal mean and variance across batches, a sufficiently large batch size is required (e.g., ScratchDet uses a batch size of 128), necessitating high-performance hardware.

[0090] The input data for neural networks typically has four dimensions: B (Batch), C (Channel), H (Height), and W (Width). During training, the amount of data that GPU memory can store, i.e., the batch size, is limited, which may be single digits in image processing tasks. To address this limitation, Normalization (GN) normalizes the data along the channel dimension. After dividing the channels into several groups, it calculates the mean and variance within each group for normalization. When calculating the mean and standard deviation, GN divides the channel dimension of each feature map into G groups, resulting in C / G channels in each group. It then calculates the mean and standard deviation for pixels belonging to the subdivided channels. Each group of channels is normally normalized independently using its corresponding parameters, so GN's operation is unaffected by the batch size and its accuracy is more stable than Batch Normalization (BN).

[0091] Meanwhile, as explained above, depthwise separable convolution and inverse residual modules were used sequentially during feature extraction to improve the feature extraction capability of the backbone network in the feature extraction part. This section will now provide a detailed explanation of this aspect. Most corn husk curl images are relative in size (…). Here, wbbox and hbbox represent the width and height of the bounding box, respectively, and wimg and himg represent the width and height of the image, respectively. For small objects less than 0.2, the backbone network used for feature extraction needs stronger extraction capabilities, typically achieved by using more complex models. However, to avoid gradient vanishing, ensure training from scratch, reduce downsampling times, maximize the receptive field of deep feature maps, and improve small object detection, the network must be as simple as possible. This is inherently contradictory. Using Depthwise Separable Convolution (DWConv) and Inverse Residual Modules (IRes) achieves a degree of balance between these two approaches. DWConv separates the steps of multiplying each convolution kernel with each feature map and then summing them. First, it performs a positional multiplication convolution along the channel dimension (with the channel unchanged), then convolves the result of the first step with a 1×1 convolution (Pointwise conv, PW), as shown in the figure below. By adjusting the number of 1×1 convolutions to change the channels, the computational cost of DWConv is approximately 1 / (kernelsize)2 of that of ordinary convolution, with a loss of only 1% precision.

[0092] The residual module can effectively reuse previous data features. As shown in the figure, its input is compressed using a 1×1 convolution, then features are extracted using a 3×3 convolution, and finally the number of channels is increased using a 1×1 convolution. At the same time, the input and output are added again, forming a data flow graph of "compression-convolution-expansion" like an hourglass. This allows the convolutional layers to focus on learning the residuals between the input and output. Directly applying DWConv to the residual module does not improve performance because DWConv's feature extraction capability is limited by the number of input channels. However, the data flow graph of the inverse residual module is "expansion-convolution-compression", similar to the shape of a spindle. Expansion is performed before the convolution operation to ensure feature extraction capability.

[0093] Meanwhile, as described above, the technical solution of this application adds cross-level data flow to the original BiFPN module of the Efficientdet target detection model to improve network performance. This will be explained in detail below.

[0094] Feature fusion fully utilizes feature map information at different resolutions, such as Figure 8As shown in section (a), good detection results for multi-scale targets can be achieved. During the bottom-up forward propagation of the network, semantic information increases with the number of downsampling iterations, while positional information gradually decreases. Although deeper feature maps possess more semantic information, their resolution is lower. After five downsampling iterations, a 32×32 pixel object in the original image is reduced to only 1×1 pixels. Therefore, deeper feature maps have low detection accuracy for small-sized targets.

[0095] The efficient bidirectional cross-scale connectivity and weighted feature fusion (BiFPN) used by EfficientDet are shown in the figure. Figure 8 As shown in Part (b), the following techniques are used to improve performance: ① fusion of features along two paths, top-down and bottom-up; ② ignoring nodes with only one input and adding skip connections to lightweight the network; ③ learning input features for automatic weighted fusion. To more fully utilize semantic and positional information at different levels, improvements are made to BiFPN (see...). Figure 8 As shown in section (c), cross-level data flow has been added, such as Figure 8 As shown in section (c), the cross-level data flow means that for any input node in the first layer of the original BiFPN module, a receiving node is determined in the second layer of the original BiFPN module to receive the input information of that input node. Then, the previous receiving node adjacent to the receiving node is determined in the second layer of the original BiFPN module, and a cross-level data flow is added between the input node and the previous receiving node. Experiments have shown that this improves network performance. Pi represents the feature map in the backbone network with a resolution of (1 / 2i) of the input image. By selecting the inverse residual module as the basic structure of the backbone network and using data normalization to optimize the gradient stability during the optimization process, the number of downsampling times of the backbone network is reduced, allowing the network to be decoupled from pre-training, and ultimately achieving end-to-end target detection.

[0096] The intelligent sprinkler irrigation system technology requires the integration of a corn leaf curling model for implementation. The irrigation system technology consists of three parts. The first part is the corn leaf curling model technology, which uses the model to determine whether large-scale leaf curling has occurred. To rule out leaf curling caused by intense midday sunlight, feedback is provided once daily at 2 PM. If the result is leaf curling, feedback is provided again at 7 PM. If the feedback is still positive, the system proceeds to the second part, which is a soil moisture sensor. This part is designed to rule out corn leaf curling caused by diseases. When the soil moisture sensor value is close to the wilting coefficient, it indicates insufficient soil moisture content. At this time, the corn shows both large-scale leaf curling and low soil moisture content, indicating a severe drought. The system then proceeds to the third part, where a drought alarm is transmitted to the backend system. After receiving the notification, the operator remotely opens the solenoid valve switch in the system to replenish the irrigation water. When the soil moisture sensor value is close to the field capacity, the system will receive another alarm, at which point the operator can remotely close the solenoid valve switch.

[0097] (1) The soil moisture sensor is equipped with a 4G card, and the information transmission system also relies on this method. Its working principle is to use the electromagnetic pulse principle and measure the apparent dielectric constant (ε) of the soil based on the propagation frequency of electromagnetic waves in the medium, thereby obtaining the soil volumetric water content (θv). The FDR has the advantages of being simple, safe, fast, accurate, continuous at fixed points, automated, wide range, and requiring little calibration. The tubular soil moisture sensor combined in this invention adopts a layered observation structure. A temperature observation point is set on the ground, and a soil temperature and humidity measurement point is set every 10cm in the underground soil to observe the soil temperature and humidity within the corresponding range. The ground part of the sensor is equipped with photovoltaic panels for solar power supply. Its materials are also sturdy and durable. After installation, no other operation is required, making it more suitable for farmers and herdsmen.

[0098] Soil moisture sensors are already available on the market, and their installation process is simple, mainly consisting of the following steps:

[0099] ① Installation time and site selection: Install the equipment after crop sowing; the installation location should be on flat terrain; under full irrigation conditions, prioritize areas with less water as monitoring locations; under partial irrigation conditions, select humid areas as monitoring locations; select locations where crop growth is balanced and can represent the growth of the vast majority of crops; understand the root distribution of the monitored crop, and generally select locations close to the crop's water-absorbing roots.

[0100] ② Drill a hole, take the drill bit, handle, and support rod of the soil drill. After completion, place the soil drill vertically on the ground, hold the handle firmly with both hands and slowly rotate it clockwise. Take the soil drill out of the hole and put it in a basin. Use a tool to collect the drilled soil into the basin for mixing with mud. Put the sensor into the drilled hole.

[0101] ③ Mix the mud: Remove the stones, roots, and other non-soluble parts from the soil extracted by the soil drill, add an appropriate amount of water, and stir into a thick mud.

[0102] ④ Grouting Installation: Slowly pour grout into the hole, filling it to approximately half its depth; adjust the amount as needed based on actual conditions. Slowly insert the sensor into the hole, rotating it gently in one direction while pressing it down. Excessive speed may prevent air bubbles from being completely expelled. Once the sensor is installed to the correct depth, some grout will overflow around the device, indicating grouting is complete; at this point, the sensor installation depth should be flush with the hole opening.

[0103] ⑤ Photovoltaic panel installation: The photovoltaic panel should be installed as far away from the sensor as possible, generally more than 50cm away, but not exceeding the length of the power cord. The photovoltaic panel should face the sun, with as few obstructions as possible in front. To fix it, simply insert the photovoltaic panel bracket into the selected location. Secure the photovoltaic panel to the bracket, aligning the four holes in the center of the panel with the four holes on the bracket, and then tighten the screws. When connecting the sensor and the photovoltaic panel, first connect the panel to the terminals on the bracket and screw them together; second, connect the solar interface of the device. Pull up the top cover of the device; the solar interface is on the side opposite the power switch. Align the other end of the bracket's power cord with the interface, insert it, and tighten the bolts. This completes the installation of the photovoltaic panel.

[0104] (2) Develop a supporting system platform / App. The uploading of corn leaf curling information and soil moisture sensor data, as well as subsequent remote solenoid valve control, all require a receiving / operating system platform. To facilitate operation, this invention is used in conjunction with an App.

[0105] The steps for developing an app are as follows:

[0106] ① Organize the complete requirements, improve the unreasonable parts, and generate the basic thinking framework.

[0107] ② Use the professional design tool Axure to draw sketches of the App to determine the general design direction of the App interface, and then refine the interface later.

[0108] ③ The front-end and back-end will gradually implement various functions.

[0109] ④ Based on the functions and features of this invention, organize the project framework.

[0110] ⑤ Complete the API interface design and interface development.

[0111] ⑥ Refine the details and conduct software testing, then launch the App.

[0112] (3) The remote irrigation system relies on the electromagnetic valve controller. The electromagnetic valve controller is equipped with a Lora module and works with the Lora gateway to achieve remote control. The 4G signal is first transmitted to the gateway, and the gateway is associated with the sub-devices below. The advantage of this is that the network requirements are further reduced. The Lora gateway has its own transmission protocol. As long as the 4G signal can be transmitted to the gateway, the network management control sub-devices no longer need network conditions. The electromagnetic valve controller is a mature product on the market and can be purchased and used directly.

[0113] Based on the same idea, embodiments of this specification also provide an apparatus corresponding to the above method, which may include:

[0114] Metal support frame;

[0115] Several rollers are provided below the metal support frame; a rotatable spherical camera with 4G signal transmission function is provided on the crossbeam of the metal support frame, and the trained corn leaf curling detection model as described in claim 1 is deployed in the spherical camera.

[0116] The metal support frame is equipped with a cable that is connected to a 220V power supply. The cable then passes through the power cable channel of the sprinkler machine and extends along the main water conveyance beam of the sprinkler to the camera, where it is connected to the camera.

[0117] The above-described apparatus and system embodiments correspond to the method embodiments and have the same technical effects. For detailed descriptions, please refer to the method embodiments. The apparatus embodiments are derived from the method embodiments; detailed descriptions can be found in the method embodiments section, and will not be repeated here.

[0118] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0119] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring, identifying and handling drought based on corn rolled leaf, characterized in that, The method comprises: Collecting corn leaf images of a detection area at a first time point and a second time point by using a rotatable spherical camera with 4G signal transmission function, wherein the first time point ranges from 12:00 to 14:00 of a day, and the second time point ranges from 17:00 to 19:00 of the day; Using the trained corn leaf rolling detection model to determine whether the corn leaves in the corn leaf images at the first time point and the second time point are in a rolling state, respectively, to obtain a determination result; If the determination result shows that the corn leaves in the corn leaf image at the first time point are in an unrolling state, the corn in the detection area is not under drought stress; if the determination result shows that the corn leaves in the corn leaf image at the first time point are in a rolling state and the corn leaves in the corn leaf image at the second time point are in a rolling state, detecting a value of a soil moisture sensor arranged in the soil of the detection area, if the value shows that the soil water content in the detection area is insufficient, the corn in the detection area is under a first degree of drought stress, and the soil of the detection area is irrigated; if the determination result shows that the corn leaves in the corn leaf image at the first time point are in a rolling state and the corn leaves in the corn leaf image at the second time point are in an unrolling state, the corn in the detection area is under a second degree of drought stress; the first degree is greater than the second degree. The training process of the trained corn leaf rolling detection model specifically comprises: Improving the Efficientdet target detection model, and using the improved Efficientdet target detection model as the corn leaf rolling detection model, wherein the improvement aspects include: Using a depth separable convolution and an inverted residual module in sequence in the feature extraction process to improve the feature extraction capability of the backbone network of the feature extraction part; Adding a cross-level data flow in the original BiFPN module of the Efficientdet target detection model to improve the network performance; Using group normalization method to normalize the data in the channel dimension during the training process, dividing the channels into several groups, calculating the mean and variance in each group for normalization, thereby preventing the phenomenon of gradient dispersion during the model training process and making the model difficult to converge; Pre-collecting a certain number of corn leaf images at the first time point and the second time point; dividing the corn leaf images into three parts of a training set, a test set and a validation set according to a predetermined ratio; Using the Pseudo-Labelling method to iteratively train the corn leaf rolling model to obtain the trained corn leaf rolling detection model.

2. The method for monitoring, identifying and handling drought according to claim 1, characterized in that, The first time point is 14:00 of a day, and the second time point is 19:00 of the day.

3. The method of claim 1, wherein the corn leaf roll monitoring is used to identify and address drought conditions. The corn leaf images are divided into three parts of a training set, a test set and a validation set in a ratio of 7:2:

1.

4. The method of claim 1, wherein the corn leaf roll monitoring is used to identify and address drought conditions. Rotating, scaling, affine transformation and noise adding are used for data enhancement to enhance the generalization ability of the trained corn leaf rolling detection model.

5. The method of claim 1, wherein the corn leaf roll monitoring is used to identify and address drought conditions. The soil in the to-be-detected area is irrigated and watered, and specifically includes the following steps. The system remotely controls the control switch installed on the sprinkler valve to remotely open the valve, irrigates and waters the soil in the to-be-detected area; when the value of the soil water content reaches the field water holding capacity, the system receives a corresponding alarm notification, and closes the valve through the system remote control switch.

6. The method of claim 1, wherein the corn leaf roll monitoring is used to identify and address drought conditions. The cross-level data flow representation, for any one of the first layer input nodes of the original BiFPN module, determines a receiving node in the second layer input node of the original BiFPN module that receives input information of the any one input node, then determines a previous receiving node adjacent to the receiving node in the second layer input node, and adds a cross-level data flow between the any one input node and the previous receiving node.

7. A device for identifying and handling drought based on corn rolled leaf monitoring, characterized in that, The device comprises: A metal support framework; A plurality of rollers are arranged below the metal support framework; a rotatable spherical camera with 4G signal transmission function is arranged on the crossbeam of the metal support framework, and the trained corn leaf rolling detection model in claim 1 is arranged in the middle of the spherical camera; A cable is arranged on the metal support framework, the cable is connected with a 220V power supply, then penetrates along a power cable channel of the sprinkler, extends to the camera along a main water supply crossbeam of the sprinkler, and is connected with the camera.

Images