Airport remote control tower multi-source information fusion monitoring equipment

Abstract

The invention relates to the technical field of airport monitoring, in particular to airport remote control tower multi-source information fusion monitoring equipment, which comprises a video monitoring module, a meteorological information module, a communication module, a data processing module and a display module, and is characterized in that the data processing module comprises an image enhancement module, a video splicing module, a target detection module and a data fusion module; the image enhancement module is used for enhancing images based on an improved multi-scale Retinex algorithm, the video stitching module is used for extracting feature points based on an SURF algorithm and stitching adjacent images, the target detection module is used for detecting targets in the images based on a YOLOv5 target detection algorithm, and the data fusion module is used for fusing multi-source data based on an improved random forest and inputting abnormal condition classification. According to the invention, through the multi-source information fusion and intelligent identification and tracking technology, the controller can master the operation condition of the airport more quickly and accurately, and the control efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of airport monitoring, in particular to a multi-source information fusion monitoring device for a remote control tower of an airport. Background Art

[0002] In today's era of rapid development of the aviation industry, the safe and efficient operation of airports is of vital importance; traditional airport tower monitoring mainly relies on a single source of information, such as radar monitoring, visual observation, etc., which can no longer meet the growing demand for aviation transportation and the complex and changing operating environment requirements.

[0003] With the continuous advancement of science and technology, various advanced monitoring technologies have emerged, but these technologies often operate independently and fail to achieve effective information fusion; for example, although traditional radars can monitor the position and speed of aircraft, they may be interfered with in severe weather conditions, and the information obtained about the specific status of the aircraft is limited; and although optical monitoring equipment can provide more intuitive image information, it is limited by observation distance and weather conditions.

[0004] In this context, there is an urgent need for an airport remote tower monitoring device that can fuse multi-source information; this device should be able to integrate data from different monitoring technologies, such as radar information, optical images, meteorological data, communication signals, etc., and through advanced information fusion technology, achieve comprehensive, accurate and real-time monitoring of the airport and surrounding airspace, and improve the safety and efficiency of airport operations. Summary of the invention

[0005] The purpose of the present invention is to provide an airport remote tower multi-source information fusion monitoring device to solve the problems raised in the above background technology.

[0006] In order to solve the above technical problems, the purpose of the present invention is to provide an airport remote tower multi-source information fusion monitoring equipment, including a video monitoring module, a meteorological information module, a communication module, a data processing module and a display module. The video monitoring module and the meteorological information module are both connected to the communication module by signal, and the communication module, the data processing module and the display module are connected by signal in sequence. The data processing module is used to process image data and fuse multi-source data to output abnormal situation classification. The data processing module includes an image enhancement module, a video splicing module, a target detection module and a data fusion module that are connected by signal in sequence. The image enhancement module enhances the image based on an improved multi-scale Retinex algorithm, the video splicing module extracts feature points based on a SURF algorithm and splices adjacent pictures, the target detection module detects targets in the spliced ​​picture based on a YOLOv5 target detection algorithm, and the data fusion module integrates multi-source data based on an improved random forest.

[0007] As a further improvement of the present technical solution, the video surveillance module includes multiple digital cameras, the video surveillance module is used to acquire video data, the communication module includes a voice communication and data communication system, the communication module is used to transmit data to the data processing module, the display module includes a screen display system and an operation console, and the display module is used to display data and remote control.

[0008] As a further improvement of the present technical solution, the meteorological information module is used to obtain real-time meteorological data, and the real-time meteorological data includes wind direction, wind speed and visibility.

[0009] As a further improvement of the technical solution, the steps of enhancing the image based on the improved multi-scale Retinex algorithm in the image enhancement module are as follows:

[0010] S1. Let the input image be I(x, y), the reflection image be R(x, y), and the illumination image be L(x, y), then:

[0011] I(x,y)=R(x,y)L(x,y);

[0012] S2. Select multiple scale parameters. For each scale parameter, use the bilateral filter function to estimate the illumination component of the logarithmic image, then:

[0013]

[0014]

[0015]

[0016]

[0017] Among them, r(x, y) is the logarithmic domain reflection image, n is the number of selected scale parameters, m is the corresponding serial number of the selected scale parameter, and w m is the weight coefficient corresponding to the mth selected scale parameter, g m (x, y) is the bilateral filter function under the mth selected scale parameter, * is the convolution operation, S x，y The center pixel I(X c ,y c ) is the domain range, I(i,j) is the pixel point in the domain, w s (i, j) is the spatial similarity factor, w r (i, j) is the brightness similarity factor, σ s and σ r All are scale factors;

[0018] S3. Calculate the reflection component, then log R (x, y) = log I (x, y) - log L (x, y), and reconstruct the enhanced image.

[0019] As a further improvement of the technical solution, the specific steps of the video stitching module extracting feature points and stitching adjacent pictures based on the SURF algorithm are as follows:

[0020] A1. Determine similar regions. Divide the edge regions from the four edges of the image respectively. Select the two edge regions with the largest structural similarity index as the splicing regions. The edge regions of the two images are p and q respectively. The size of the edge region is M×N. Then:

[0021]

[0022]

[0023]

[0024]

[0025] C1=(K1L) 2

[0026] C2=(K2L) 2

[0027] Among them, μ represents the average brightness of the image, e(i, j) represents the pixel value of the image at the coordinate (i, j), and σ 2 represents the variance of the image, σ pq represents the covariance of edge region p and edge region q, p(i, j) represents the pixel value of edge region p at coordinate (i, j), μ p represents the average brightness of the edge region p, q(i, j) represents the pixel value of the edge region q at the coordinate (i, j), μ q represents the average brightness of the edge area q, SSIM pq represents the structural similarity index of edge region p and edge region q, C1 and C2 are both stability coefficients, represents the variance of the edge region p, represents the variance of the edge region q, K1 and K2 are both constants, and L represents the dynamic range of pixel values;

[0028] A2. Generate Hessian matrix to provide stable edge feature points for feature point extraction;

[0029] A3. Construct a multi-scale spatial pyramid using box filters;

[0030] A4, feature point positioning, using Harr wavelet transform to assign feature point directions, generate feature point descriptors and feature point matching;

[0031] A5. Use the random sampling consensus algorithm to remove incorrect matching pairs and obtain the optimal transformation matrix. Transform one of the images to align the two images, and then fuse the two aligned images.

[0032] As a further improvement of the present technical solution, the target detection module uses the image data spliced ​​by the video splicing module as input data, and detects humans and animals based on the YOLOv5 target detection algorithm.

[0033] As a further improvement of the technical solution, the specific steps of the data fusion module fusing multi-source information to determine abnormal situations are:

[0034] B1. The multi-source information includes the output data of the meteorological information module and the target detection module, and the multi-source information is used as input data;

[0035] B2. Based on the improved random forest algorithm, the abnormal situation is judged. The m input data are divided into a training set and a validation set. Boostrap is used to extract c samples from the training set, and n features are randomly extracted to build a decision tree. The optimal partitioning attribute is selected according to the Gini index, and the partitioning is continued until the depth D of the decision tree is reached or all samples of the node are of the same category. N decision trees are constructed. The calculation formula of the Gini index is:

[0036]

[0037] Among them, G s represents the Gini coefficient of node s on the decision tree, R represents the total number of categories, and p rs Indicates the proportion of samples belonging to the rth class at node s on the decision tree;

[0038] B3, based on the improved particle swarm algorithm to optimize the number of decision trees N, the number of extracted features n and the depth of decision trees D in the random forest;

[0039] B4. Use the trained random forest to classify abnormal situations.

[0040] As a further improvement of the technical solution, the specific steps of optimizing the parameters in the random forest in B3 are:

[0041] B3.1, initialize parameters, set the population size, initialize the position and speed of each particle, set the maximum number of iterations, determine the spatial dimension, the spatial dimension is three, and the position of the particle represents the number of decision trees N, the number of extracted features n, and the depth of the decision tree D in the random forest;

[0042] B3.2. Calculate the fitness function of each particle;

[0043] B3.3, Update speed and position, then:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] Among them, w represents the inertia weight, w start represents the initial value of inertia weight, w end represents the final value of the inertia weight, t represents the current number of iterations, and t max represents the maximum number of iterations, c1 represents the individual learning factor, represents the initial value of the individual learning factor, represents the final value of the individual learning factor, c2 represents the group learning factor, represents the initial value of the group learning factor, represents the final value of the group learning factor, represents the velocity of the i-th particle in the t+1th iteration, represents the velocity of the ith particle in the tth iteration, r1 and r2 are both random numbers in [0, 1]. represents the best position that the ith particle has passed at the tth iteration, gbest represents the global optimal position, represents the position of the i-th particle at the t-th iteration, represents the position of the i-th particle at the t+1th iteration;

[0050] B3.4. Repeat B3.2 and B3.3 and stop when the maximum number of iterations is reached.

[0051] As a further improvement of this technical solution, the fitness function in B3.2 is:

[0052]

[0053] Among them, Accuracy represents the accuracy, g c represents the number of correctly classified samples, and G represents the total number of samples.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows: in the airport remote tower multi-source information fusion monitoring equipment, through multi-source information fusion and intelligent recognition and tracking technology, the controller can grasp the operation of the airport more quickly and accurately, improve the control efficiency, and also reduce the controller's visual blind spots and misjudgment rate through panoramic monitoring, real-time enhancement and intelligent recognition and other technical means, thereby improving control safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is the overall structure diagram of the present invention;

[0056] Figure 2 It is a structural diagram of the data processing module in the present invention;

[0057] The meaning of each mark in the figure is:

[0058] 1. Video surveillance module;

[0059] 2. Weather information module;

[0060] 3. Communication module;

[0061] 4. Data processing module; 40. Image enhancement module; 41. Video stitching module; 42. Target detection module; 43. Data fusion module;

[0062] 5. Display module. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] like Figure 1 and Figure 2As shown, the purpose of this embodiment is to provide an airport remote tower multi-source information fusion monitoring equipment, including a video monitoring module 1, a meteorological information module 2, a communication module 3, a data processing module 4 and a display module 5. The video monitoring module 1 and the meteorological information module 2 are both connected to the communication module 3 by signal. The communication module 3, the data processing module 4 and the display module 5 are connected in sequence by signal. The data processing module 4 is used to process image data and fuse multi-source data to output abnormal situation classification. The data processing module 4 includes an image enhancement module 40, a video splicing module 41, a target detection module 42 and a data fusion module 43 that are connected in sequence by signal. The image enhancement module 40 enhances the image based on an improved multi-scale Retinex algorithm, the video splicing module 41 extracts feature points based on a SURF algorithm and splices adjacent images, the target detection module 42 detects targets in the spliced ​​image based on a YOLOv5 target detection algorithm, and the data fusion module 43 fuses multi-source data based on an improved random forest and inputs abnormal situation classification, which can organically fuse information from multiple sources, provide controllers with comprehensive monitoring and command means, and improve control efficiency and safety.

[0065] The video monitoring module 1 includes multiple digital cameras, which are used to obtain video data to achieve panoramic monitoring of the airport scene and display of local details. The communication module 3 includes voice communication and data communication systems. The communication module 3 is used to transmit data to the data processing module 4 to ensure real-time communication between the remote tower and the airport site and other control units. The display module 5 includes a screen display system and an operation console. The display module 5 is used to display data and remote control, providing controllers with an intuitive and convenient monitoring and command interface.

[0066] The meteorological information module 2 is used to obtain real-time meteorological data, including wind direction, wind speed and visibility, to provide decision support for controllers.

[0067] The steps of the image enhancement module 40 for enhancing the image based on the improved multi-scale Retinex algorithm are as follows:

[0068] S1. Let the input image be I(x, y), the reflection image be R(x, y), and the illumination image be L(x, y), then:

[0069] I(x,y)=R(x,y)L(x,y);

[0070] S2. Select multiple scale parameters. For each scale parameter, use the bilateral filter function to estimate the illumination component of the logarithmic image, then:

[0071]

[0072]

[0073]

[0074]

[0075] Among them, r(x, y) is the logarithmic domain reflection image, n is the number of selected scale parameters, m is the corresponding serial number of the selected scale parameter, and w m is the weight coefficient corresponding to the mth selected scale parameter, g m (x, y) is the bilateral filter function under the mth selected scale parameter, * is the convolution operation, S x，y is the center pixel I(x c ,y c ) is the domain range, I(i,j) is the pixel point in the domain, w s (i, j) is the spatial similarity factor, w r (i, j) is the brightness similarity factor, σ s and σ r Both are scale factors. Bilateral filtering considers the similarity of pixel values ​​while considering the pixel spatial distance. When optimizing multi-scale Retinex, it can effectively retain the edge details in the image, making the edge of the enhanced image clearer and sharper.

[0076] S3. Calculate the reflection component, then logR(x, y)=logI(x, y)-logL(x, y), and reconstruct the enhanced image. The multi-scale Retinex algorithm can effectively enhance the contrast of the image, so that the dark details in the image can be clearly presented, and the bright area will not be overexposed, thereby making the overall visual effect of the image more vivid. At the same time, the algorithm can well maintain the color constancy of the image during the processing process, and will not cause color distortion due to the enhancement operation, ensuring that the color of the image is real and natural.

[0077] The specific steps of the video stitching module 41 extracting feature points and stitching adjacent pictures based on the SURF algorithm are as follows:

[0078] A1. Determine similar areas. Divide the edge areas from the four edges of the image respectively. Select the two edge areas with the largest structural similarity index as the stitching area. Comprehensively consider the similarity of the images from three aspects: brightness, contrast and structure. Compared with the method based only on pixel differences, it is more in line with the human visual system's perception of image quality, can accurately reflect the overall structural information of the image, has high sensitivity to subtle changes and distortions in the image, and shows good adaptability and stability when evaluating different types of images. At the same time, the calculation is relatively simple and efficient, which is easy to apply in various image processing and analysis scenarios. It provides reliable quantitative indicators for tasks such as image quality assessment and image matching, avoids detecting feature points in the entire image, and reduces the amount of calculation. The edge areas of the two images are p and q respectively, and the size of the edge area is M×N, then:

[0079]

[0080]

[0081]

[0082]

[0083] C1=(K1L) 2

[0084] C2=(K2L) 2

[0085] Among them, μ represents the average brightness of the image, e(i, j) represents the pixel value of the image at the coordinate (i, j), and σ 2 represents the variance of the image, σ pq represents the covariance of edge region p and edge region q, p(i, j) represents the pixel value of edge region p at coordinate (i, j), μ p represents the average brightness of the edge region p, q(i, j) represents the pixel value of the edge region q at the coordinate (i, j), μ q represents the average brightness of the edge area q, SSIM pq represents the structural similarity index of edge region p and edge region q, C1 and C2 are both stability coefficients, represents the variance of the edge region p, represents the variance of the edge region q, K1 and K2 are both constants, and L represents the dynamic range of pixel values;

[0086] A2. Generate Hessian matrix to provide stable edge feature points for feature point extraction. Hessian matrix can effectively detect local structural changes in images and has high sensitivity to features such as edges and corners. By generating Hessian matrix, stable feature points in images can be found quickly and accurately. These feature points have good invariance under different image transformations, providing a reliable basis for subsequent feature point matching and image stitching.

[0087] A3. Use box filters to construct multi-scale spatial pyramids. The use of box filters makes the construction of multi-scale space more efficient, can quickly detect feature points at different scales, and improve the running speed of the algorithm;

[0088] A4. Feature point positioning: using Harr wavelet transform to assign directions to feature points, generate feature point descriptors and match feature points. Harr wavelet transform assigns directions to feature points, which can make feature points rotationally invariant, that is, no matter how the image is rotated, the descriptors of feature points can remain relatively stable. Generating feature point descriptors can convert feature points into a set of representative numerical vectors, which is convenient for matching feature points. Feature point matching is to find corresponding feature point pairs in the two images to be spliced, providing a basis for subsequent image alignment.

[0089] A5. Use the random sampling consistency algorithm to remove incorrect matching pairs and obtain the optimal transformation matrix. Transform one of the images to align the two images, and fuse the two aligned images. Through the random sampling consistency algorithm, reliable feature point matching pairs can be screened out to obtain the optimal transformation matrix. Transform one of the images to align it with the other image to ensure that the two images can be accurately spliced ​​together in space. Finally, fuse the two aligned images to make the transition at the splicing point natural, avoid obvious splicing marks, and improve the quality of the spliced ​​image.

[0090] The target detection module 42 uses the image data spliced ​​by the video splicing module 41 as input data, and detects people and animals based on the YOLOv5 target detection algorithm, and promptly alerts the controller when an abnormal situation occurs.

[0091] The specific steps of the data fusion module 43 fusing multi-source information to determine abnormal situations are:

[0092] B1, the multi-source information includes the output data of the meteorological information module 2 and the target detection module 42, and the multi-source information may also include information from multiple sources such as radar and ADS-B, and the multi-source information is used as input data;

[0093] B2. Based on the improved random forest algorithm, the abnormal situation is judged. The m input data are divided into a training set and a validation set. Boostrap is used to extract c samples from the training set, and n features are randomly extracted to build a decision tree. The optimal partitioning attribute is selected according to the Gini index, and the partitioning is continued until the depth D of the decision tree is reached or all samples of the node are of the same category. N decision trees are constructed. The calculation formula of the Gini index is:

[0094]

[0095] Among them, G s represents the Gini coefficient of node s on the decision tree, R represents the total number of categories, and p rs It represents the proportion of samples belonging to the rth class on node s on the decision tree. Random forest is used to fuse multi-source information. Random forest can reduce the error that may be caused by a single information source by integrating the prediction results of multiple decision trees. Different information sources often have their own advantages and limitations. For example, an information source may perform well in some cases, but not well in other scenarios. Through the fusion of random forest, the advantages of multiple information sources can complement each other, thereby reducing the overall prediction error. The fusion of multi-source information also makes random forest more robust to noise and outliers in the data. When a certain information source is interfered by noise, other information sources can play a certain correction role, thereby maintaining the stability of the model;

[0096] B3, based on the improved particle swarm algorithm to optimize the number of decision trees N, the number of extracted features n and the depth of decision trees D in the random forest;

[0097] The specific steps for optimizing the parameters in random forest in B3 are:

[0098] B3.1, initialize parameters, set the population size, initialize the position and speed of each particle, set the maximum number of iterations, determine the spatial dimension, the spatial dimension is three, and the position of the particle represents the number of decision trees N, the number of extracted features n, and the depth of the decision tree D in the random forest;

[0099] B3.2. Calculate the fitness function of each particle;

[0100] The fitness function in B3.2 is:

[0101]

[0102] Among them, Accuracy represents the accuracy, g c represents the number of correctly classified samples, and G represents the total number of samples;

[0103] B3.3, Update speed and position, then:

[0104]

[0105]

[0106]

[0107]

[0108]

[0109] Among them, w represents the inertia weight, w start represents the initial value of inertia weight, w end represents the final value of the inertia weight, t represents the current number of iterations, and t max represents the maximum number of iterations, c1 represents the individual learning factor, represents the initial value of the individual learning factor, represents the final value of the individual learning factor, c2 represents the group learning factor, represents the initial value of the group learning factor, represents the final value of the group learning factor, represents the velocity of the i-th particle in the t+1th iteration, represents the velocity of the ith particle in the tth iteration, r1 and r2 are both random numbers in [0, 1]. represents the best position that the ith particle has passed at the tth iteration, gbest represents the global optimal position, represents the position of the i-th particle at the t-th iteration, It represents the position of the i-th particle at the t+1th iteration. Through the linear decrease of the inertia weight and the adaptive adjustment of the learning factor, the search behavior of the particle is dynamically adjusted during the operation of the algorithm, which helps to improve the search efficiency and convergence speed of the algorithm.

[0110] B3.4, repeat B3.2 and B3.3, and stop when the maximum number of iterations is reached;

[0111] B4. Use the trained random forest to classify abnormal situations.

[0112] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An airport remote tower multi-source information fusion monitoring device, characterized in that: The invention comprises a video monitoring module (1), a meteorological information module (2), a communication module (3), a data processing module (4) and a display module (5), wherein the video monitoring module (1) and the meteorological information module (2) are both connected to the communication module (3) by signals, and the communication module (3), the data processing module (4) and the display module (5) are connected to each other by signals in sequence, the data processing module (4) is used to process image data and fuse multi-source data to output abnormal situation classification, the data processing module (4) comprises an image enhancement module (40), a video splicing module (41), a target detection module (42) and a data fusion module (43) which are connected to each other by signals in sequence, the image enhancement module (40) enhances images based on an improved multi-scale Retinex algorithm, the video splicing module (41) extracts feature points based on a SURF algorithm and splices adjacent images, the target detection module (42) detects targets in the spliced ​​images based on a YOLOv5 target detection algorithm, and the data fusion module (43) fuses multi-source data based on an improved random forest and inputs abnormal situation classification.

2. The airport remote tower multi-source information fusion monitoring equipment according to claim 1 is characterized by: The video monitoring module (1) comprises a multi-channel digital camera, the video monitoring module (1) is used to acquire video data, the communication module (3) comprises a voice communication and data communication system, the communication module (3) is used to transmit data to the data processing module (4), the display module (5) comprises a screen display system and an operation console, and the display module (5) is used to display data and remotely control.

3. The airport remote tower multi-source information fusion monitoring equipment according to claim 1 is characterized by: The meteorological information module (2) is used to obtain real-time meteorological data, which includes wind direction, wind speed and visibility.

4. The airport remote tower multi-source information fusion monitoring equipment according to claim 1 is characterized in that: The steps of enhancing the image by the image enhancement module (40) based on the improved multi-scale Retinex algorithm are as follows: S1. Let the input image be I(x,y), the reflection image be R(x,y), and the illumination image be L(x,y), then: I(x,y)=R(x,y)L(x,y); S2. Select multiple scale parameters. For each scale parameter, use the bilateral filter function to estimate the illumination component of the logarithmic image, then: Among them, r(x,y) is the logarithmic domain reflection image, n is the number of selected scale parameters, m is the corresponding serial number of the selected scale parameter, and w m is the weight coefficient corresponding to the mth selected scale parameter, g m (x, y) is the bilateral filter function under the mth selected scale parameter, * is the convolution operation, S x,y is the center pixel I(x c ,y c ) is the domain range, I(i,j) is the pixel point in the domain, w s (i,j) is the spatial similarity factor, w r (i,j) is the brightness similarity factor, σ s and σ r All are scale factors; S3. Calculate the reflection component, then logR(x,y)=logI(x,y)-logL(x,y), and reconstruct the enhanced image.

5. The airport remote tower multi-source information fusion monitoring equipment according to claim 1 is characterized in that: The specific steps of the video splicing module (41) extracting feature points based on the SURF algorithm and splicing adjacent pictures are: A1. Determine similar regions. Divide the edge regions from the four edges of the image respectively. Select the two edge regions with the largest structural similarity index as the splicing regions. The edge regions of the two images are p and q respectively. The size of the edge region is M×N. Then: C1=(K1L) 2 C2=(K2L) 2 Among them, μ represents the average brightness of the image, e(i,j) represents the pixel value of the image at coordinate (i,j), σ 2 represents the variance of the image, σ pq represents the covariance of edge region p and edge region q, p(i,j) represents the pixel value of edge region p at coordinate (i,j), μ p represents the average brightness of the edge region p, q(i,j) represents the pixel value of the edge region q at the coordinate (i,j), μ q represents the average brightness of the edge area q, SSIM pq represents the structural similarity index of edge region p and edge region q, C1 and C2 are both stability coefficients, represents the variance of the edge region p, represents the variance of the edge region q, K1 and K2 are both constants, and L represents the dynamic range of pixel values; A2. Generate Hessian matrix to provide stable edge feature points for feature point extraction; A3. Construct a multi-scale spatial pyramid using box filters; A4, feature point positioning, using Harr wavelet transform to assign feature point directions, generate feature point descriptors and feature point matching; A5. Use the random sampling consensus algorithm to remove incorrect matching pairs and obtain the optimal transformation matrix. Transform one of the images to align the two images, and then fuse the two aligned images.

6. The airport remote tower multi-source information fusion monitoring equipment according to claim 1 is characterized by: The target detection module (42) uses the image data spliced ​​by the video splicing module (41) as input data, and detects humans and animals based on the YOLOv5 target detection algorithm.

7. The airport remote tower multi-source information fusion monitoring equipment according to claim 1 is characterized in that: The specific steps of the data fusion module (43) fusing multi-source information to determine abnormal situations are: B1, the multi-source information includes the output data of the meteorological information module (2) and the target detection module (42), and the multi-source information is used as input data; B2. Based on the improved random forest algorithm to judge abnormal situations, m input data are divided into training set and validation set. Boostrap is used to extract c samples from the training set, and n features are randomly extracted to build a decision tree. The optimal partitioning attribute is selected according to the Gini index, and the partitioning is continued until the depth D of the decision tree is reached or all samples of the node are of the same category. N decision trees are constructed. The calculation formula of the Gini index is: Among them, G s represents the Gini coefficient of node s on the decision tree, R represents the total number of categories, and p rs Indicates the proportion of samples belonging to the rth class at node s on the decision tree; B3, optimize the number of decision trees N, the number of extracted features n and the depth of decision trees D in random forest based on improved particle swarm algorithm; B4. Use the trained random forest to classify abnormal situations.

8. The airport remote tower multi-source information fusion monitoring equipment according to claim 7 is characterized by: The specific steps of optimizing the parameters in the random forest in B3 are: B3.1, initialize parameters, set the population size, initialize the position and speed of each particle, set the maximum number of iterations, determine the spatial dimension, the spatial dimension is three, and the position of the particle represents the number of decision trees N, the number of extracted features n, and the depth of the decision tree D in the random forest; B3.

2. Calculate the fitness function of each particle; B3.3, Update speed and position, then: Among them, w represents the inertia weight, w start represents the initial value of inertia weight, w end represents the final value of the inertia weight, t represents the current number of iterations, and t max represents the maximum number of iterations, c1 represents the individual learning factor, represents the initial value of the individual learning factor, represents the final value of the individual learning factor, c2 represents the group learning factor, represents the initial value of the group learning factor, represents the final value of the group learning factor, represents the velocity of the i-th particle in the t+1th iteration, represents the velocity of the ith particle in the tth iteration, r1 and r2 are both random numbers in [0,1]. represents the best position that the ith particle has passed at the tth iteration, gbest represents the global optimal position, represents the position of the i-th particle at the t-th iteration, represents the position of the i-th particle at the t+1th iteration; B3.

4. Repeat B3.2 and B3.3 and stop when the maximum number of iterations is reached.

9. The airport remote tower multi-source information fusion monitoring equipment according to claim 8 is characterized by: The fitness function in B3.2 is: Among them, Accuracy represents the accuracy, g c represents the number of correctly classified samples, and G represents the total number of samples.

Images