Target monitoring method and system based on unmanned aerial vehicle cruise

Abstract

The invention provides a target monitoring method and system based on unmanned aerial vehicle cruise, and relates to the technical field of computers. The method comprises the following steps: firstly, acquiring an RGB crop image, a crop spectral image and fuselage attitude data formed by acquiring a target area by a target unmanned aerial vehicle during cruising; secondly, based on red, green and blue wave bands in the crop spectral image, channel enhancement processing is carried out on the RGB crop image, and an enhanced crop image vector is formed; then, carrying out global fusion processing on the red edge wave band in the crop spectrum image and the enhanced crop image vector to form a crop global semantic vector; further, on the basis of semantic information of the fuselage attitude data, performing gating screening processing on the crop global semantic vector to form a target crop semantic vector; and finally, analyzing crop monitoring data based on the target crop semantic vector. Based on the above content, the problem that the reliability of crop monitoring is relatively low in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a target monitoring method and system based on unmanned aerial vehicle (UAV) cruise. Background Art

[0002] With the rapid development of agricultural modernization, drone technology is increasingly being used in agricultural monitoring. Drones, equipped with sensors (such as image, temperature, and humidity sensors), can monitor farmland environments and crop growth, providing crucial data support for crop health management, pest and disease control, and yield forecasting. However, existing crop monitoring technologies still face significant challenges in practical applications. For example, traditional crop monitoring methods typically rely on a single type of sensor data, such as RGB images, which cannot fully reflect crop status. Furthermore, threshold comparison and analysis of temperature and humidity data are often required, making it difficult to reliably and effectively monitor crop status. Summary of the Invention

[0003] In view of this, the object of the present invention is to provide a target monitoring method and system based on drone cruising, so as to improve the problem of relatively low reliability of crop monitoring in the prior art.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A target monitoring method based on UAV cruise, comprising: Acquire an RGB crop image and a crop spectral image formed by the target UAV capturing the target area during cruising, and acquire body posture data of the target UAV when the image is formed; performing channel enhancement processing on the RGB crop image based on semantic information of red, green and blue bands in the crop spectral image to form an enhanced crop image vector; performing a global fusion process on the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector to form a crop global semantic vector; Based on the semantic information of the fuselage posture data, gated screening is performed on the crop global semantic vector to form a target crop semantic vector; Crop monitoring data of the target area is analyzed based on the target crop semantic vector, wherein the crop monitoring data is used to reflect the degree of plant diseases and insect pests in the target area.

[0005] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of globally fusing the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector to form a global semantic vector of the crop includes: For each image pixel in the crop spectral image, determining a crop state parameter corresponding to the image pixel based on a difference between an upper limit spectral parameter and a lower limit spectral parameter of a red edge band of the image pixel, wherein the crop state parameter is positively correlated with the difference; constructing a crop state parameter matrix based on the crop state parameter corresponding to each image pixel in the crop spectral image; The crop state parameter matrix and the enhanced crop image vector are globally fused to form a crop global semantic vector, wherein the global fusion processing includes a first semantic mining and a second semantic mining, the first semantic mining is used to mine the crop state parameter matrix, and the second semantic mining is used to fuse the enhanced crop image vector in the process of mining the results of the first semantic mining.

[0006] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of globally fusing the crop state parameter matrix and the enhanced crop image vector to form a crop global semantic vector includes: performing a first semantic mining on the crop state parameter matrix to form crop state vectors at multiple scales, wherein each of the crop state vectors is formed by performing a semantic mining on a compression parameter matrix of a corresponding scale formed by performing a compression operation; Performing a second semantic mining on the internal correlation vector of the a-1th scale and the enhanced crop image vector to form an external correlation vector of the ath scale, wherein the internal correlation vector of the 0th scale is a crop state vector with the smallest scale among the crop state vectors of the multiple scales; The external association vector of the a-th scale and the crop state vector of the a-th scale are subjected to cross-attention processing and vector expansion operations to form an internal association vector of the a-th scale, wherein the internal association vector of the last scale is used as the crop global semantic vector.

[0007] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of performing a first semantic mining on the crop state parameter matrix to form crop state vectors at multiple scales includes: Step a, selecting any crop state parameter from the crop state parameter matrix as the first crop state parameter in the target parameter set, selecting a new crop state parameter based on the principle of maximizing the distance from each crop state parameter in the current target parameter set, and adding the new crop state parameter to the current target parameter set to form a new target parameter set, until the number of crop state parameters in the new target parameter set reaches a target number, wherein after performing step a multiple times, multiple target parameter sets are formed; Step b: determining a central crop state parameter in each target parameter set, and, for each central crop state parameter, determining a neighborhood parameter matrix centered on the central crop state parameter in the crop state parameter matrix; Step c: performing a pooling operation on each of the neighborhood parameter matrices to form a pooling matrix corresponding to each of the neighborhood parameter matrices, and concatenating each of the pooling matrices to form a crop state vector of the largest scale. After constructing a new crop state parameter matrix based on each central crop state parameter, steps a, b, and c are re-executed based on the new crop state parameter matrix and the new target quantity to determine a crop state vector of the second largest scale.

[0008] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of performing a second semantic mining on the internal correlation vector of the a-1th scale and the enhanced crop image vector to form an external correlation vector of the ath scale includes: Based on the vector size of the internal correlation vector at the a-1th scale, screening the vector parameters in the enhanced crop image vector to form an enhanced crop image vector at the a-1th scale; The internal correlation vector of the a-1th scale and the enhanced crop image vector of the a-1th scale are cross-attention processed to form an external correlation vector of the a-th scale.

[0009] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of screening the vector parameters in the enhanced crop image vector based on the vector size of the internal correlation vector at the a-1th scale to form the enhanced crop image vector at the a-1th scale includes: For each vector parameter in the enhanced crop image vector, determining a local image vector corresponding to the vector parameter based on the vector size of the internal correlation vector at the a-1th scale, with the vector parameter as the center, and determining a vector similarity between the local image vector and the internal correlation vector at the a-1th scale; screening a plurality of target vector parameters from the enhanced crop image vector based on the vector similarity and the vector size of the internal correlation vector at the a-1th scale; For each of the target vector parameters, performing a pooling operation on the local image vector corresponding to the target vector parameter to form a pooled image vector corresponding to the target vector parameter; Based on the pooled image vector corresponding to each of the target vector parameters, an enhanced crop image vector of the a-1th scale is formed by combining them.

[0010] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of performing channel enhancement processing on the RGB crop image based on the semantic information of the red, green and blue bands in the crop spectral image to form an enhanced crop image vector includes: determining a red band matrix based on spectral parameters of a red band in the crop spectral image, determining a red channel matrix based on color values of a red channel in the RGB crop image, and performing channel enhancement processing on the red channel matrix based on the red band matrix to form a red channel increase vector; determining a green band matrix based on spectral parameters of a green band in the crop spectral image, determining a green channel matrix based on color values of a green channel in the RGB crop image, and performing channel enhancement processing on the green channel matrix based on the green band matrix to form a green channel increase vector; determining a blue band matrix based on spectral parameters of a blue band in the crop spectral image, determining a blue channel matrix based on color values of a blue channel in the RGB crop image, and performing channel enhancement processing on the blue channel matrix based on the blue band matrix to form a blue channel increase vector; A mean calculation is performed on the red channel added vector, the green channel added vector, and the blue channel added vector to obtain an enhanced crop image vector.

[0011] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruise, the steps of determining a red band matrix based on spectral parameters of the red band in the crop spectral image, determining a red channel matrix based on color values of the red channel in the RGB crop image, and performing channel enhancement processing on the red channel matrix based on the red band matrix to form a red channel increase vector include: For each image pixel in the crop spectral image, performing a mean calculation based on the spectral parameters of the image pixel in the red band to obtain a mean of the red spectral parameters of the image pixel, and constructing a red band matrix based on the mean of the red spectral parameters of each image pixel; constructing a red channel matrix based on the color value of the red channel in the RGB crop image; Performing depth convolution processing on the red band matrix and the red channel matrix respectively to form a red band depth vector and a red channel depth vector; Based on the cross-attention mechanism, the red band depth vector and the red channel depth vector are cross-fused to form a red channel added vector, wherein, in the cross-fusion process, the query vector comes from the red band depth vector, and the key vector and the value vector both come from the red channel depth vector.

[0012] In a preferred embodiment of the present invention, in the above-mentioned target monitoring method based on drone cruising, the step of performing gated screening on the crop global semantic vector based on the semantic information of the fuselage posture data to form the target crop semantic vector includes: Performing semantic space conversion processing on the fuselage posture data to form a fuselage posture vector; Performing linear mapping processing on the fuselage attitude vector to form a fuselage attitude linear vector; performing nonlinear activation on the fuselage attitude linear vector to form a fuselage attitude activation vector; The fuselage posture activation vector is used as a gating parameter to perform gated filtering processing on the crop global semantic vector to form a target crop semantic vector, wherein any vector parameter in the target crop semantic vector is equal to the product of the fuselage posture activation vector and the vector parameter at the corresponding position in the crop global semantic vector.

[0013] On the basis of the above, the present invention also provides a target monitoring system based on drone cruise, including: a memory for storing computer programs; a processor connected to the memory, for executing the computer programs stored in the memory to implement the above-mentioned target monitoring method based on drone cruise.

[0014] The present invention provides a target monitoring method and system based on drone cruise. First, RGB crop images, crop spectral images, and fuselage posture data are acquired by a target drone during cruise. Second, channel enhancement processing is performed on the RGB crop image based on the red, green, and blue bands in the crop spectral image to form an enhanced crop image vector. Then, the red edge band in the crop spectral image and the enhanced crop image vector are globally fused to form a global crop semantic vector. Further, based on the semantic information of the fuselage posture data, the global crop semantic vector is gated and filtered to form a target crop semantic vector. Finally, crop monitoring data is analyzed based on the target crop semantic vector. Based on the above, on the one hand, because the semantic information of different bands in the crop spectral image is processed separately, the processing accuracy is higher. Specifically, channel enhancement processing is performed on the RGB crop image based on the semantic information of the red, green, and blue bands, thereby improving the semantic representation accuracy of the three color channels. Moreover, because the semantic information of the red edge band has a relatively relevant representation effect on the health status of the crop, the corresponding global fusion can form rich semantic information in the semantic representation of the health status dimension. On the other hand, by using the semantic information of the drone's posture data as the basis for gating and filtering, interference caused by changes in the drone's posture can be effectively removed. Therefore, by leveraging these two advantages, the semantic vectors of target crops can be generated with both semantic richness and semantic representation accuracy, thereby improving the reliability of the crop monitoring data derived from this analysis and addressing the relatively low reliability of crop monitoring in existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings.

[0016] Figure 1 This is a structural block diagram of an electronic device provided by an embodiment of the present invention.

[0017] Figure 2 A schematic flow chart of a target monitoring method based on drone cruising provided in an embodiment of the present invention.

[0018] Figure 3 A schematic diagram of the fusion of RGB crop images and crop spectral images provided by an embodiment of the present invention.

[0019] Figure 4 A schematic diagram of mapping the semantic information of the red edge band into a vector space provided by an embodiment of the present invention.

[0020] Figure 5 A schematic diagram of global fusion processing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, an embodiment of the present invention provides a target monitoring system based on drone cruising. The target monitoring system based on drone cruising may include a memory and a processor.

[0023] Specifically, the memory and the processor are electrically connected, directly or indirectly, to enable data transmission or interaction. For example, the memory and the processor may be electrically connected via one or more communication buses or signal lines. The processor is configured to execute executable computer programs stored in the memory, such as software functional modules and computer programs included in a drone cruise-based target monitoring device, to implement the drone cruise-based target monitoring method provided in an embodiment of the present invention.

[0024] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0025] Optionally, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0026] Optionally, the target monitoring device based on drone cruising may include: An image acquisition module is used to acquire RGB crop images and crop spectral images formed by the target UAV capturing the target area during cruising, and to acquire the body posture data of the target UAV when the image is formed; a channel enhancement module, configured to perform channel enhancement processing on the RGB crop image based on semantic information of red, green, and blue bands in the crop spectral image to form an enhanced crop image vector; a global fusion module, configured to perform global fusion processing on the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector to form a crop global semantic vector; a gating and filtering module, configured to perform gating and filtering processing on the crop global semantic vector based on the semantic information of the fuselage posture data to form a target crop semantic vector; The semantic analysis module is used to analyze the crop monitoring data of the target area based on the target crop semantic vector, wherein the crop monitoring data is used to reflect the degree of plant diseases and insect pests in the target area.

[0027] I understand. Figure 1 The structure shown is for illustration only. The target monitoring system based on UAV cruise can also include Figure 1 More or fewer components than shown, or with Figure 1 The different configurations shown, for example, may also include a communication unit for exchanging information with other devices (such as an RGB image sensor, a spectral sensor, etc. carried by a drone).

[0028] Combine Figure 2 Embodiments of the present invention further provide a drone-based cruise target monitoring method applicable to the aforementioned drone-based cruise target monitoring system. The method steps defined in the process associated with the drone-based cruise target monitoring method can be implemented by the drone-based cruise target monitoring system (hereinafter referred to as the target monitoring system).

[0029] The following will Figure 2The specific process shown is explained in detail.

[0030] Step S110 , obtaining an RGB crop image and a crop spectral image formed by the target UAV collecting data of the target area during cruising, and obtaining the body posture data of the target UAV when the image is formed.

[0031] In an embodiment of the present invention, the target monitoring system can capture RGB crop images and crop spectral images of the target area generated by the target drone during navigation, as well as the drone's body posture data during image generation. The RGB crop images can be captured by the target drone's onboard RGB camera, and the crop spectral images can be captured by the drone's onboard spectral imaging equipment. Furthermore, the body posture data can include flight altitude, speed, attitude angles (pitch, pitch, and yaw), etc., such as "flight altitude: 10 meters, speed: 0.5 meters per second, pitch: 0 degrees, yaw: 5 degrees."

[0032] Step S120 : performing channel enhancement processing on the RGB crop image based on semantic information of the red, green, and blue bands in the crop spectral image to form an enhanced crop image vector.

[0033] In an embodiment of the present invention, after acquiring the crop spectral image and the RGB crop image, the target monitoring system can perform channel enhancement processing on the RGB crop image based on the semantic information of the red, green, and blue bands in the crop spectral image to form an enhanced crop image vector. In other words, the semantic information of the red, green, and blue bands can be extracted from the crop spectral image, allowing the semantic information of the red, green, and blue channels in the RGB crop image to be enhanced. Specifically, the semantic information of the red channel is enhanced based on the semantic information of the red band, the semantic information of the green channel is enhanced based on the semantic information of the green band, and the semantic information of the blue channel is enhanced based on the semantic information of the blue band. This ensures the accuracy of the semantic enhancement.

[0034] Step S130 , globally fusing the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector to form a global crop semantic vector.

[0035] In an embodiment of the present invention, after forming the enhanced crop image vector, the target monitoring system may further globally fuse the semantic information of the red edge band in the crop spectral image with the enhanced crop image vector to form a global crop semantic vector. It should be noted that in the red edge band, the spectral reflectance of plant leaves typically exhibits a steep downward trend due to chlorophyll's strong absorption of red light. The slope of this downward trend reflects the chlorophyll content and the level of photosynthetic activity. In other words, if the semantic information in the red edge band indicates that the corresponding downward trend has become flat, it may reflect a low level of photosynthetic activity, indicating possible exposure to pests and diseases. The semantic information in the RGB crop image can, to a certain extent, reflect the color and morphology of the crop, which also has a certain correlation with pests and diseases. Therefore, the semantic information in the red edge band and the enhanced crop image vector can be fused to form semantic information that can characterize the crop's status from multiple angles, namely, the global crop semantic vector.

[0036] Step S140 : performing gated screening processing on the crop global semantic vector based on the semantic information of the fuselage posture data to form a target crop semantic vector.

[0037] In an embodiment of the present invention, after forming the global crop semantic vector, the target monitoring system can perform gated filtering on the global crop semantic vector based on the semantic information of the drone posture data to form a target crop semantic vector. Specifically, because there may be subtle differences between the RGB crop images and crop spectral image representations captured by drones in different postures through corresponding equipment, to avoid interference caused by these differences, the global crop semantic vector can be gated and filtered based on the semantic information of the drone posture data, thereby obtaining a target crop semantic vector with higher semantic representation accuracy.

[0038] Step S150 : Analyze the crop monitoring data of the target area based on the target crop semantic vector.

[0039] In an embodiment of the present invention, after forming the target crop semantic vector, the target monitoring system may analyze the target crop semantic vector to generate crop monitoring data for the target area. The crop monitoring data is used to reflect the severity of crop pests and diseases in the target area. For example, the crop monitoring data may be represented by a numerical value ranging from 0 to 1, specifically, a higher numerical value indicates a higher severity of pests and diseases, and a lower numerical value indicates a lower severity of pests and diseases.

[0040] For the above steps, for example: First, data collection can be performed: UAV A-1, equipped with high-resolution equipment, flies over the rapeseed field, captures RGB images and spectral images, and records flight attitude data; Secondly, data processing can be performed: RGB and spectral images are enhanced, global fusion is performed to form a global semantic vector, and gated filtering is performed based on the posture data to remove posture interference; Finally, we can analyze the results: the analysis shows that the degree of disease and insect pests in rapeseed is mild.

[0041] Based on the above content, on the one hand, since the semantic information of different bands in the crop spectral image is processed separately, the processing accuracy is higher. Specifically, the RGB crop image is channel enhanced through the semantic information of the red, green and blue bands, so as to improve the accuracy of the semantic representation in the three color channels. Moreover, since the semantic information of the red edge band has a more relevant representation effect on the health status of the crop, through the corresponding global fusion, rich semantic information can be formed in the semantic representation of the health status dimension (combined with Figure 3 On the other hand, by using the semantic information of the drone's posture data as the basis for gating and filtering, interference caused by changes in the drone's posture can be effectively removed. Therefore, by leveraging these two advantages, the generated semantic vector for the target crop can achieve both semantic richness and semantic representation accuracy, thereby improving the reliability of the crop monitoring data derived from this analysis and addressing the relatively low reliability of crop monitoring in existing technologies.

[0042] In the first part, it is necessary to further explain step S110 that the specific method of obtaining the RGB crop image, the crop spectral image, and the body posture data is not limited and can be selected accordingly according to actual needs.

[0043] For example, in a specific embodiment, the target UAV may directly establish a communication connection with the target monitoring system, so that the RGB crop image, the crop spectral image, and the body posture data may be directly acquired.

[0044] For example, in another specific embodiment, the RGB crop image, crop spectral image, and fuselage posture data of the target UAV can be first stored in a corresponding database. Then, when crop status analysis is required, the target monitoring system obtains the RGB crop image, the crop spectral image, and the fuselage posture data from the database.

[0045] In the second part, it is necessary to further explain step S120 , that the specific method of performing channel enhancement processing on the RGB crop image is not limited and can be selected according to actual needs.

[0046] For example, in a specific embodiment, based on the semantic information of the three bands of red, green and blue in the crop spectral image, channel enhancement processing can be performed on the semantic information of the three color channels of red, green and blue in the RGB crop image respectively. Then, the semantic vectors after the enhancement of the three color channels of red, green and blue can be spliced to form an enhanced crop image vector.

[0047] For example, in another specific embodiment, in order to take into account the accuracy of the enhancement processing and the amount of data of the enhanced crop image vector formed, the above-mentioned step S120 may further include step S121, step S122, step S123 and step S124, and the specific content of each step is described as follows.

[0048] Step S121: Determine a red band matrix based on spectral parameters of the red band in the crop spectral image, determine a red channel matrix based on color values of the red channel in the RGB crop image, and perform channel enhancement processing on the red channel matrix based on the red band matrix to form a red channel increase vector.

[0049] In an embodiment of the present invention, a red band matrix can be determined based on the spectral parameters of the red band in the crop spectral image, and a red channel matrix can be determined based on the color values of the red channel in the RGB crop image. Furthermore, channel enhancement processing can be performed on the red channel matrix based on the red band matrix to form a red channel increase vector. That is, when performing enhancement processing on the red channel alone, considering the correlation between the spectral information of each image pixel in the crop spectral image and the correlation between the channel color values of each image pixel in the RGB crop image, a corresponding matrix can be formed to represent this correlation. This matrix can then better capture the aforementioned correlation during the channel enhancement process, thereby improving the semantic representation accuracy of the resulting red channel increase vector.

[0050] Step S122: determining a green band matrix based on spectral parameters of the green band in the crop spectral image, determining a green channel matrix based on color values of the green channel in the RGB crop image, and performing channel enhancement processing on the green channel matrix based on the green band matrix to form a green channel increase vector.

[0051] In an embodiment of the present invention, a green band matrix can be determined based on the spectral parameters of the green band in the crop spectral image, and a green channel matrix can be determined based on the color values of the green channel in the RGB crop image. Furthermore, channel enhancement processing can be performed on the green channel matrix based on the green band matrix to form a green channel increase vector. That is, when performing enhancement processing on the green channel alone, considering the correlation between the spectral information of each image pixel in the crop spectral image and the correlation between the channel color values of each image pixel in the RGB crop image, a corresponding matrix can be formed to represent this correlation. This matrix can then better capture the aforementioned correlation during the channel enhancement process, thereby improving the semantic representation accuracy of the resulting green channel increase vector.

[0052] Step S123 , determining a blue band matrix based on spectral parameters of the blue band in the crop spectral image, determining a blue channel matrix based on color values of the blue channel in the RGB crop image, and performing channel enhancement processing on the blue channel matrix based on the blue band matrix to form a blue channel increase vector.

[0053] In an embodiment of the present invention, a blue band matrix can be determined based on the spectral parameters of the blue band in the crop spectral image, and a blue channel matrix can be determined based on the color values of the blue channel in the RGB crop image. Furthermore, channel enhancement processing can be performed on the blue channel matrix based on the blue band matrix to form a blue channel increase vector. That is, when performing enhancement processing on the blue channel alone, considering the correlation between the spectral information of each image pixel in the crop spectral image and the correlation between the channel color values of each image pixel in the RGB crop image, a corresponding matrix can be formed to represent this correlation. This matrix can then better capture the aforementioned correlation during the channel enhancement process, thereby improving the semantic representation accuracy of the resulting blue channel increase vector.

[0054] Step S124 , performing mean calculation on the red channel added vector, the green channel added vector, and the blue channel added vector to obtain an enhanced crop image vector.

[0055] In an embodiment of the present invention, after obtaining the red channel increase vector, the green channel increase vector, and the blue channel increase vector, a mean calculation may be performed on the red channel increase vector, the green channel increase vector, and the blue channel increase vector to obtain an enhanced crop image vector. In other words, the mean of the three vector parameters at the same position in the red channel increase vector, the green channel increase vector, and the blue channel increase vector may be calculated to obtain the enhanced crop image vector.

[0056] It is understandable that, in the above-mentioned steps S121, S122, and S123, the channel enhancement processing may be performed in the same manner. For example, in a specific embodiment, the above-mentioned step S121 may further include the following contents (step S122 and step S123 are similar): In the first step, for each image pixel in the crop spectral image, a mean is calculated based on the spectral parameters of the image pixel in the red band (since the red band has multiple wavelengths, it corresponds to multiple spectral parameters, and a mean calculation can be performed), to obtain the mean of the red spectral parameters of the image pixel. Furthermore, based on the mean of the red spectral parameters of each image pixel, a red band matrix is constructed. For example, the spectral parameters of each image pixel in the crop spectral image in the red band can be replaced with the corresponding mean of the red spectral parameters to form a red band matrix, i.e., the mean is used as the representative value, or the median value, etc., can also be used as the representative value; In the second step, a red channel matrix may be constructed based on the color value of the red channel in the RGB crop image; In the third step, deep convolution processing can be performed on the red band matrix and the red channel matrix respectively to form a red band depth vector and a red channel depth vector; that is, deep convolution processing is performed respectively to extract the spectral deep semantic information of the red band from the red band matrix, and to extract the color deep semantic information of the red channel from the red channel matrix; wherein, deep mining processing can be implemented by two different convolution units, each convolution unit can include multiple convolution layers to gradually improve the depth of the captured semantic features, and each convolution layer can include one or more convolution kernels, for example, the size of the convolution kernel can be 3*3 or 5*5, etc. It should be noted that the size of the red band depth vector and the red channel depth vector is the same; In the fourth step, based on the cross-attention mechanism, the red band depth vector and the red channel depth vector can be cross-fused to form a red channel added vector. In the cross-fusion process, the query vector comes from the red band depth vector, and the key vector and the value vector both come from the red channel depth vector. The specific fusion process can refer to the relevant existing technology.

[0057] In the third part, it is necessary to further explain step S130 that the specific method of globally fusing the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector is not limited and can be selected according to actual needs.

[0058] For example, in a specific embodiment, after the semantic information of the red edge band in the crop spectral image is mapped to the vector space, considering that the semantic space between the semantic vector formed by the mapping and the enhanced crop image vector is different, the semantic space can be converted to the semantic vector formed by the mapping, and then the converted semantic vector and the enhanced crop image vector can be averaged or spliced to form a global semantic vector of the crop.

[0059] Among them, for each image pixel in the crop spectral image, the spectral parameters of the image pixel in the red edge band can be sampled. If the number of samples is n, a semantic vector of A*B*n can be formed, where A*B is the number of image pixels. Then, channel convolution processing can be performed (for example, the size of the convolution kernel is 1*1*n, and then, for each image pixel in the semantic vector of A*B*n, the n vector parameters corresponding to the n image pixels are weighted summed based on the parameters of the convolution kernel to form the vector parameters after convolution), thereby forming the semantic vector of A*B (combined with Figure 4 Then, the semantic vector and a transformation matrix may be multiplied, and the multiplication result may be added to a transposed parameter to obtain a transformed semantic vector.

[0060] For example, in another specific embodiment, in order to ensure that the formed global semantic vector of the crop can fully characterize the health status of the crop in the photosynthesis dimension, the above-mentioned step S130 can further include step S131, step S132 and step S133, the specific contents of which are as follows.

[0061] Step S131 : for each image pixel in the crop spectral image, determining a crop state parameter corresponding to the image pixel based on a difference between an upper limit spectral parameter and a lower limit spectral parameter of a red edge band of the image pixel.

[0062] In this embodiment of the present invention, for each image pixel in the crop spectral image, a crop state parameter corresponding to the image pixel is determined based on the difference between the upper and lower spectral parameters of the red edge band of the image pixel. The crop state parameter is positively correlated with the difference. For example, the crop state parameter can be determined according to the following calculation formula: ; Wherein, E is the crop state parameter, r1 is the upper limit spectral parameter, such as the spectral reflectance corresponding to the maximum wavelength in the red edge band (such as 750 nm or 770 nm), and r2 is the lower limit spectral parameter, such as the spectral reflectance corresponding to the minimum wavelength in the red edge band (such as 700 nm).

[0063] Step S132 : constructing a crop state parameter matrix based on the crop state parameter corresponding to each image pixel in the crop spectral image.

[0064] In this embodiment of the present invention, a crop state parameter matrix can be constructed based on the crop state parameters corresponding to each image pixel in the crop spectral image. Specifically, the corresponding crop state parameters are combined according to the distribution of each image pixel in the crop spectral image, thereby forming a crop state parameter matrix with the same distribution as the crop spectral image. This crop state parameter matrix can effectively characterize the adjacent relationships between the crop state parameters of adjacent image pixels, allowing for exploration in subsequent processing.

[0065] Step S133 : globally fusing the crop state parameter matrix and the enhanced crop image vector to form a crop global semantic vector.

[0066] In an embodiment of the present invention, after constructing the crop state parameter matrix, the crop state parameter matrix and the enhanced crop image vector are globally fused to form a crop global semantic vector. The global fusion process includes a first semantic mining process and a second semantic mining process, wherein the first semantic mining process is used to mine the crop state parameter matrix, and the second semantic mining process is used to fuse the enhanced crop image vector in the process of mining the results of the first semantic mining process. That is, since the crop state parameter matrix has a direct correlation with the representation of photosynthesis intensity, it plays a relatively important role. Thus, in the process of global fusion processing, the first semantic mining process can be performed first to mine the deep potential semantic information therein, and then the second semantic mining process can be further performed. In the process of mining, the fusion of the semantic information in the enhanced crop image vector is achieved, thereby ensuring that the deep semantic information is mined and the fusion of semantic information is achieved, thereby improving the richness.

[0067] Alternatively, in the above step S133, the specific manner of globally fusing the crop state parameter matrix and the enhanced crop image vector is not limited. For example, in a specific embodiment, in order to fully fuse the enhanced crop image vector on the basis of mining the deep potential semantic information in the crop state parameter matrix, the above step S133 may further include step S133a, step S133b and step S133c. The specific contents of each step are as follows (combined with Figure 5 ).

[0068] Step S133a: performing a first semantic mining on the crop state parameter matrix to form crop state vectors of multiple scales.

[0069] In an embodiment of the present invention, the crop state parameter matrix can be subjected to a first semantic mining to form a plurality of scale crop state vectors. Each of the crop state vectors is formed by semantic mining a compression parameter matrix of a corresponding scale formed by a compression operation. That is, for any scale, the crop state parameter matrix can be subjected to a compression operation to form a compression parameter matrix of the corresponding scale. Then, the compression parameter matrix can be subjected to semantic mining to form a corresponding scale crop state vector. Based on this, a plurality of scale crop state vectors can be formed, such as Figure 5 The crop state vectors at the fourth scale, the third scale, the second scale, and the first scale are shown. By compressing the crop state parameter matrix at different scales, we can capture feature information at different levels and granularities, providing a more comprehensive description of the crop state.

[0070] Step S133b: performing a second semantic mining on the internal correlation vector of the a-1th scale and the enhanced crop image vector to form an external correlation vector of the ath scale.

[0071] In an embodiment of the present invention, after forming the crop state vectors at multiple scales, the internal correlation vector at the a-1th scale and the enhanced crop image vector may be subjected to a second semantic mining to form an external correlation vector at the ath scale. The internal correlation vector at the 0th scale is the smallest scale crop state vector among the crop state vectors at the multiple scales, such as Figure 5 In addition, a is an integer greater than or equal to 1 and less than or equal to Q, wherein the specific value of Q can be configured according to actual conditions. For example, Q can be equal to the crop state vectors of the multiple scales to achieve full adaptation of the first semantic mining and the second semantic mining. In addition, as Figure 5As shown, the second semantic mining can be achieved through the external association unit.

[0072] Step S133c: performing cross-attention processing and vector expansion operations on the external correlation vector of the a-th scale and the crop state vector of the a-th scale to form an internal correlation vector of the a-th scale.

[0073] In the embodiment of the present invention, after forming the external correlation vector of the ath scale, the external correlation vector of the ath scale and the crop state vector of the ath scale can be subjected to cross attention processing and vector expansion operations to form the internal correlation vector of the ath scale. The internal correlation vector of the last scale is used as the crop global semantic vector, such as Figure 5 The internal association vector of the fourth scale in the a-th scale is used as the global semantic vector of the crop. Exemplarily, the external association vector of the a-th scale can be cross-attention processed based on the crop state vector of the a-th scale (that is, the query vector comes from the crop state vector of the a-th scale, and the key vector and the value vector both come from the external association vector of the a-th scale), and then the vector obtained by the cross-attention processing is subjected to a vector expansion operation (such as through upsampling processing to achieve an increase in size, wherein, in the process of the first semantic mining, the compression operation can be used to gradually reduce the corresponding size from the crop state vector of the fourth scale to the crop state vector of the first scale, and in the process of the second semantic mining, the vector expansion operation can be used to gradually increase the corresponding size) to form the internal association vector of the a-th scale. In addition, as Figure 5 As shown, the cross attention processing and vector expansion operation can be implemented by the internal association unit. It should be further explained that the method provided by the present invention can be implemented by a neural network model formed by training, wherein the neural network model can include a first semantic mining network and a second semantic mining network, the first semantic mining network can be used to perform the above-mentioned first semantic mining, the second semantic mining network can include multiple sub-networks, each sub-network can include an external association unit and an internal association unit, such as Figure 5 As shown, the second semantic mining network may include 4 sub-networks.

[0074] Based on the above scheme, the fusion of the enhanced crop image vectors can be achieved first through external association, and then the fusion of the crop state vectors of the corresponding scales can be achieved through internal association. In this way, the full fusion of the enhanced crop image vectors can be achieved through multiple scales.

[0075] Optionally, in step S133a, the specific manner of performing the first semantic mining on the crop state parameter matrix is not limited. For example, in a specific embodiment, in order to achieve feature extraction at different scales while ensuring the diversity of mined semantic information and avoiding one-sided information (i.e., the crop state vector at each scale can represent the semantic information of the crop state parameter matrix as a whole), thereby improving the comprehensiveness and accuracy of feature mining, step S133a may further include the following: In a first step, any crop state parameter is selected from the crop state parameter matrix as the first crop state parameter in the target parameter set. Based on the principle of maximizing the distance from each crop state parameter in the current target parameter set (i.e., the distance between each crop state parameter in the current target parameter set has the maximum value, where the distance may refer to the distance between distribution positions in the matrix), a new crop state parameter is selected and added to the current target parameter set to form a new target parameter set, until the number of crop state parameters in the new target parameter set reaches a target number (e.g., the target number is equal to half of the crop state parameters in the crop state parameter matrix). After executing the first step multiple times, multiple target parameter sets are formed, and the selected crop state parameter is different for every two executions of the first step. In a second step, a central crop state parameter is determined in each target parameter set (for example, a crop state parameter having the smallest mean distance from other crop state parameters can be used as the central crop state parameter). Furthermore, for each central crop state parameter, a neighborhood parameter matrix centered on the central crop state parameter is determined in the crop state parameter matrix. The size of the neighborhood parameter matrix can be selected based on actual needs. For example, when high accuracy is required, the size can be larger, and when a small amount of computational data is required, the size can be smaller. Specifically, the size of the neighborhood parameter matrix can be 3*3, 5*5, or 7*7, etc. In the third step, a pooling operation is performed on each of the neighborhood parameter matrices to form a pooling matrix corresponding to each of the neighborhood parameter matrices (for example, a mean or maximum size can be performed, that is, the mean or maximum value of each crop state parameter in the neighborhood parameter matrix is determined. In this case, the size of the pooling matrix can be 1*1), and each pooling matrix is spliced to form a crop state vector of the largest scale. After a new crop state parameter matrix is constructed based on each central crop state parameter, the first, second, and third steps are re-executed based on the new crop state parameter matrix and a new target number (e.g., equal to 1 / 2 of the target number) to determine a crop state vector of the second largest scale. In this way, a crop state vector of the third largest scale (the corresponding new target number can be 1 / 4 of the target number) and a crop state vector of the fourth largest scale (the corresponding new target number can be 1 / 8 of the target number) can be determined in sequence.

[0076] It should be noted that the maximum scale crop state vector can be Figure 5 The crop state vector of the fourth scale in the , the crop state vector of the second largest scale can be Figure 5 The crop state vector of the third scale in the , the crop state vector of the third largest scale can be Figure 5 The crop state vector of the second scale in the , the crop state vector of the fourth scale can be Figure 5 The crop state vector of the first scale in .

[0077] Optionally, in the above-mentioned step S133b, the specific method of performing the second semantic mining on the internal correlation vector of the a-1th scale and the enhanced crop image vector is not limited. For example, in a specific embodiment, considering that the internal correlation vectors of each scale have different sizes, and the enhanced crop image vector has a fixed size, in order to achieve reliable second semantic mining, the above-mentioned step S133b may further include step b1 and step b2. The specific contents of each step are described below.

[0078] Step b1: Based on the vector size of the internal correlation vector of the a-1th scale, the vector parameters in the enhanced crop image vector are screened to form an enhanced crop image vector of the a-1th scale.

[0079] In an embodiment of the present invention, the size of the enhanced crop image vector may be the same as the size of the crop state parameter matrix. Therefore, during the fusion mining process, the vector parameters in the enhanced crop image vector may be screened based on the vector size of the internal correlation vector at the a-1th scale to form an enhanced crop image vector at the a-1th scale, that is, the vector size of the enhanced crop image vector is compressed to the vector size of the internal correlation vector at the a-1th scale, thereby obtaining an enhanced crop image vector at the a-1th scale having the corresponding vector size.

[0080] Step b2: performing cross-attention processing on the internal correlation vector of the a-1th scale and the enhanced crop image vector of the a-1th scale to form an external correlation vector of the a-th scale.

[0081] In an embodiment of the present invention, after obtaining the enhanced crop image vector of the a-1th scale, the internal correlation vector of the a-1th scale and the enhanced crop image vector of the a-1th scale can be cross-attention processed, wherein the query vector can come from the enhanced crop image vector of the a-1th scale, and the key vector and the value vector can both come from the internal correlation vector of the a-1th scale.

[0082] Optionally, in step b1 above, the specific method for screening the vector parameters in the enhanced crop image vector is not limited. For example, in a specific embodiment, in order to achieve size compression through screening and also optimize the enhanced crop image vector through screening, thereby ensuring high accuracy of semantic representation of the compressed enhanced crop image vector and avoiding semantic distortion caused by compression, step b1 above may include the following: In the first step, for each vector parameter in the enhanced crop image vector, the local image vector corresponding to the vector parameter can be determined based on the vector size of the internal correlation vector at the a-1th scale and centered on the vector parameter (i.e., the vector size of the local image vector is equal to the vector size of the internal correlation vector at the a-1th scale), and the vector similarity (e.g., cosine similarity) between the local image vector and the internal correlation vector at the a-1th scale can be determined. In a second step, a plurality of target vector parameters may be screened from the enhanced crop image vector based on the vector similarity and the vector size of the internal correlation vector at the a-1th scale. For example, a plurality of vector parameters with the greatest vector similarity may be screened and used as target vector parameters, wherein the number of the plurality of target vector parameters is equal to the number of vector parameters in the internal correlation vector at the a-1th scale, i.e., matches the vector size. In this manner, through similarity-based feature screening, it is ensured that the selected features are highly correlated with the corresponding internal correlation vectors, thereby enhancing the effectiveness of the features. In the third step, a pooling operation may be performed on the local image vector corresponding to each target vector parameter to form a pooled image vector corresponding to the target vector parameter. For example, the mean or maximum value of each vector parameter in the local image vector may be calculated. In this way, the vector size of the corresponding pooled image vector may be 1*1. In the fourth step, based on the pooled image vectors corresponding to each of the target vector parameters, an enhanced crop image vector of the a-1th scale may be formed by combining (concatenating).

[0083] In the fourth part, it is necessary to further explain step S140 that the specific method of performing gated screening on the crop global semantic vector is not limited and can be selected according to actual needs.

[0084] For example, in a specific embodiment, in order to better capture the global relationship between posture semantic information and crop feature semantic information, in addition to dynamically adjusting the focus on crop features according to the aircraft posture through gated screening, the above-mentioned step S140 may further include the following: The first step is to perform semantic space mapping on the fuselage posture data to obtain a posture vector; illustratively, the fuselage posture data can be embedded using a word embedding model; In the second step, the attention mechanism is used to calculate the similarity between the pose vector and the crop global semantic vector; The third step is to weight the crop semantic vectors according to the similarity to form the target crop semantic vector. In this way, the similarity can be used as the basis for gating screening.

[0085] For example, in another specific embodiment, in order to efficiently process and enhance semantic information, reduce redundant data, and improve computational efficiency, in addition to dynamically adjusting the focus on crop features according to the aircraft's posture through gated screening, step S140 may further include the following: In the first step, the fuselage posture data is subjected to semantic space conversion processing to form a fuselage posture vector. For example, the fuselage posture data can be embedded through a word embedding model (such as a Word2Vec model). For example, by performing word segmentation embedding on "the flight altitude is 10 meters...", we can obtain: {flight: [0.8, 0.5, 0.7, 0.3, 0.6, 0.2, 0.9, 0.4, 0.5, 0.8]; height: [0.6, 0.7, 0.4, 0.8, 0.2, 0.5, 0.3, 0.9, 0.7, 0.6]; height: [0.5, 0.3, 0.8, 0.2, 0.7, 0.6, 0.4, 0.5, 0.9, 0.3]; height: [0.1, 0.1, 0.7, 0.4, m:[0.2, 0.8, 0.6, 0.4, 0.1, 0.9, 0.7,0.3, 0.5, 0.6]; ...}; In a second step, a linear mapping process may be performed on the fuselage attitude vector to form a fuselage attitude linear vector, wherein the linear mapping process may be implemented by a corresponding linear mapping function, and the linear mapping function may be F(x)=Ax+b, wherein A represents a weight matrix and b represents a bias parameter; In the third step, nonlinear activation can be performed on the fuselage attitude linear vector to form a fuselage attitude activation vector, wherein the nonlinear activation can be implemented by a nonlinear activation function, and the nonlinear activation function can be a Sigmoid function, etc.; In the fourth step, the fuselage posture activation vector can be used as a gating parameter to perform gated screening on the crop global semantic vector to form a target crop semantic vector, wherein any vector parameter in the target crop semantic vector is equal to the product of the fuselage posture activation vector and the vector parameter at the corresponding position in the crop global semantic vector, that is, the fuselage posture activation vector and the crop global semantic vector are bitwise multiplied.

[0086] In the fifth part, it needs to be further explained with respect to step S150 that the specific method of analyzing the crop monitoring data of the target area based on the target crop semantic vector is not limited and can be selected accordingly according to actual needs. For example, in a specific implementation, the target crop semantic vector can be fully connected to obtain an output vector of size 1*1, and then, the output vector can be subjected to identity mapping and linear mapping to obtain a parameter value to characterize the degree of pests and diseases, namely, the crop monitoring data.

[0087] In summary, the target monitoring method and system based on drone cruise provided by the present invention first obtains RGB crop images, crop spectral images, and fuselage posture data formed by the target drone collecting data of the target area during cruise; secondly, based on the red, green, and blue bands in the crop spectral image, the RGB crop image is channel-enhanced to form an enhanced crop image vector; then, the red edge band in the crop spectral image and the enhanced crop image vector are globally fused to form a crop global semantic vector; further, based on the semantic information of the fuselage posture data, the crop global semantic vector is gated and filtered to form a target crop semantic vector; finally, crop monitoring data is analyzed based on the target crop semantic vector. Based on the above content, on the one hand, because the semantic information of different bands in the crop spectral image is processed separately, the processing accuracy is higher. Specifically, the semantic information of the red, green, and blue bands is used to enhance the RGB crop image channel, thereby improving the semantic representation accuracy in the three color channels. Moreover, because the semantic information of the red edge band has a relatively relevant representation effect on the health status of the crop, the corresponding global fusion can form rich semantic information in the semantic representation of the health status dimension. On the other hand, by using the semantic information of the drone's posture data as the basis for gating and filtering, interference caused by changes in the drone's posture can be effectively removed. Therefore, by leveraging these two advantages, the semantic vectors of target crops can be generated with both semantic richness and semantic representation accuracy, thereby improving the reliability of the crop monitoring data derived from this analysis and addressing the relatively low reliability of crop monitoring in existing technologies.

[0088] In the several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0089] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A target monitoring method based on UAV cruise, characterized in that: include: Acquire an RGB crop image and a crop spectral image formed by the target UAV capturing the target area during cruising, and acquire body posture data of the target UAV when the image is formed; performing channel enhancement processing on the RGB crop image based on semantic information of red, green and blue bands in the crop spectral image to form an enhanced crop image vector; performing a global fusion process on the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector to form a crop global semantic vector; Based on the semantic information of the fuselage posture data, gated screening is performed on the crop global semantic vector to form a target crop semantic vector; Crop monitoring data of the target area is analyzed based on the target crop semantic vector, wherein the crop monitoring data is used to reflect the degree of plant diseases and insect pests in the target area.

2. The target monitoring method based on drone cruising according to claim 1 is characterized in that: The step of globally fusing the semantic information of the red edge band in the crop spectral image and the enhanced crop image vector to form a global crop semantic vector includes: For each image pixel in the crop spectral image, determining a crop state parameter corresponding to the image pixel based on a difference between an upper limit spectral parameter and a lower limit spectral parameter of a red edge band of the image pixel, wherein the crop state parameter is positively correlated with the difference; constructing a crop state parameter matrix based on the crop state parameter corresponding to each image pixel in the crop spectral image; The crop state parameter matrix and the enhanced crop image vector are globally fused to form a crop global semantic vector, wherein the global fusion processing includes a first semantic mining and a second semantic mining, the first semantic mining is used to mine the crop state parameter matrix, and the second semantic mining is used to fuse the enhanced crop image vector in the process of mining the results of the first semantic mining.

3. The target monitoring method based on drone cruising according to claim 2 is characterized in that: The step of globally fusing the crop state parameter matrix and the enhanced crop image vector to form a crop global semantic vector includes: performing a first semantic mining on the crop state parameter matrix to form crop state vectors at multiple scales, wherein each of the crop state vectors is formed by performing a semantic mining on a compression parameter matrix of a corresponding scale formed by performing a compression operation; Performing a second semantic mining on the internal correlation vector of the a-1th scale and the enhanced crop image vector to form an external correlation vector of the ath scale, wherein the internal correlation vector of the 0th scale is a crop state vector with the smallest scale among the crop state vectors of the multiple scales; The external association vector of the a-th scale and the crop state vector of the a-th scale are subjected to cross-attention processing and vector expansion operations to form an internal association vector of the a-th scale, wherein the internal association vector of the last scale is used as the crop global semantic vector.

4. The target monitoring method based on UAV cruise according to claim 3 is characterized in that: The step of performing a first semantic mining on the crop state parameter matrix to form crop state vectors at multiple scales includes: Step a, selecting any crop state parameter from the crop state parameter matrix as the first crop state parameter in the target parameter set, selecting a new crop state parameter based on the principle of maximizing the distance from each crop state parameter in the current target parameter set, and adding the new crop state parameter to the current target parameter set to form a new target parameter set, until the number of crop state parameters in the new target parameter set reaches a target number, wherein after performing step a multiple times, multiple target parameter sets are formed; Step b: determining a central crop state parameter in each target parameter set, and, for each central crop state parameter, determining a neighborhood parameter matrix centered on the central crop state parameter in the crop state parameter matrix; Step c: performing a pooling operation on each of the neighborhood parameter matrices to form a pooling matrix corresponding to each of the neighborhood parameter matrices, and concatenating each of the pooling matrices to form a crop state vector of the largest scale. After constructing a new crop state parameter matrix based on each central crop state parameter, steps a, b, and c are re-executed based on the new crop state parameter matrix and the new target quantity to determine a crop state vector of the second largest scale.

5. The target monitoring method based on UAV cruise according to claim 3 is characterized in that: The step of performing a second semantic mining on the internal correlation vector of the a-1th scale and the enhanced crop image vector to form an external correlation vector of the ath scale includes: Based on the vector size of the internal correlation vector at the a-1th scale, screening the vector parameters in the enhanced crop image vector to form an enhanced crop image vector at the a-1th scale; The internal correlation vector of the a-1th scale and the enhanced crop image vector of the a-1th scale are cross-attention processed to form an external correlation vector of the a-th scale.

6. The target monitoring method based on UAV cruise according to claim 5 is characterized in that: The step of screening the vector parameters in the enhanced crop image vector based on the vector size of the internal correlation vector at the a-1th scale to form the enhanced crop image vector at the a-1th scale includes: For each vector parameter in the enhanced crop image vector, determining a local image vector corresponding to the vector parameter based on the vector size of the internal correlation vector at the a-1th scale, with the vector parameter as the center, and determining a vector similarity between the local image vector and the internal correlation vector at the a-1th scale; screening a plurality of target vector parameters from the enhanced crop image vector based on the vector similarity and the vector size of the internal correlation vector at the a-1th scale; For each of the target vector parameters, performing a pooling operation on the local image vector corresponding to the target vector parameter to form a pooled image vector corresponding to the target vector parameter; Based on the pooled image vector corresponding to each of the target vector parameters, an enhanced crop image vector of the a-1th scale is formed by combining them.

7. The target monitoring method based on UAV cruise according to claim 1 is characterized in that: The step of performing channel enhancement processing on the RGB crop image based on semantic information of the red, green, and blue bands in the crop spectral image to form an enhanced crop image vector includes: determining a red band matrix based on spectral parameters of a red band in the crop spectral image, determining a red channel matrix based on color values of a red channel in the RGB crop image, and performing channel enhancement processing on the red channel matrix based on the red band matrix to form a red channel increase vector; determining a green band matrix based on spectral parameters of a green band in the crop spectral image, determining a green channel matrix based on color values of a green channel in the RGB crop image, and performing channel enhancement processing on the green channel matrix based on the green band matrix to form a green channel increase vector; determining a blue band matrix based on spectral parameters of a blue band in the crop spectral image, determining a blue channel matrix based on color values of a blue channel in the RGB crop image, and performing channel enhancement processing on the blue channel matrix based on the blue band matrix to form a blue channel increase vector; A mean calculation is performed on the red channel added vector, the green channel added vector, and the blue channel added vector to obtain an enhanced crop image vector.

8. The target monitoring method based on UAV cruise according to claim 7 is characterized in that: The steps of determining a red band matrix based on spectral parameters of the red band in the crop spectral image, determining a red channel matrix based on color values of the red channel in the RGB crop image, and performing channel enhancement processing on the red channel matrix based on the red band matrix to form a red channel increase vector include: For each image pixel in the crop spectral image, performing a mean calculation based on the spectral parameters of the image pixel in the red band to obtain a mean of the red spectral parameters of the image pixel, and constructing a red band matrix based on the mean of the red spectral parameters of each image pixel; constructing a red channel matrix based on the color value of the red channel in the RGB crop image; Performing depth convolution processing on the red band matrix and the red channel matrix respectively to form a red band depth vector and a red channel depth vector; Based on the cross-attention mechanism, the red band depth vector and the red channel depth vector are cross-fused to form a red channel added vector, wherein, in the cross-fusion process, the query vector comes from the red band depth vector, and the key vector and the value vector both come from the red channel depth vector.

9. The target monitoring method based on drone cruising according to any one of claims 1 to 8, characterized in that: The step of performing gated screening processing on the crop global semantic vector based on the semantic information of the fuselage posture data to form a target crop semantic vector includes: Performing semantic space conversion processing on the fuselage posture data to form a fuselage posture vector; Performing linear mapping processing on the fuselage attitude vector to form a fuselage attitude linear vector; performing nonlinear activation on the fuselage attitude linear vector to form a fuselage attitude activation vector; The fuselage posture activation vector is used as a gating parameter to perform gated filtering processing on the crop global semantic vector to form a target crop semantic vector, wherein any vector parameter in the target crop semantic vector is equal to the product of the fuselage posture activation vector and the vector parameter at the corresponding position in the crop global semantic vector.

10. A target monitoring system based on drone cruising, characterized in that: include: memory for storing computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the target monitoring method based on drone cruising as described in any one of claims 1 to 9.

Images