Recycling resource detecting method based on drone image using artificial intelligence

Abstract

A method for detecting recyclable resources based on drone imagery using artificial intelligence is disclosed. A method for detecting recyclable resources based on drone imagery using artificial intelligence according to an embodiment of the present invention is a method for detecting recyclable resources based on drone imagery using artificial intelligence performed by a server, comprising: receiving a photograph in which a drone captures a predetermined area; performing spatial information processing on the photograph to generate spatial information data; inputting the spatial information data into a pre-trained artificial intelligence model to detect recyclable resources on the spatial information data and outputting a detection area; assigning actual coordinates to the detection area; and reflecting the spatial information data to the detection area to calculate the volume of the recyclable resources. According to the present invention, a method for detecting recyclable resources based on drone imagery using artificial intelligence is provided, which generates spatial information from a drone's captured imagery, detects recyclable resources in the captured imagery using artificial intelligence, and calculates the scale of the recyclable resources based on the spatial information to enable collection.

Description

Technology Field

[0001] The present invention relates to a method for detecting recyclable resources, and more specifically, to a method for detecting recyclable resources based on drone imagery using artificial intelligence, wherein spatial information is generated from a drone's captured image, recyclable resources in the captured image are detected using artificial intelligence, and the location and scale of the recyclable resources are calculated based on the spatial information to enable collection. Background Technology

[0003] The amount of unmanaged waste illegally dumped is increasing every year. A significant portion of this waste includes recyclable resources. Currently, managing, collecting, and processing this waste requires substantial manpower and costs.

[0004] In particular, in the case of unauthorized, abandoned, and unmanaged waste, it is difficult to identify easily because it is left in places that are hard for people to access, or it is practically difficult to locate the site of the illegal dumping due to the vastness of the area.

[0005] Therefore, there is a need to develop technology that enables the detection and collection of illegally dumped waste using unmanned aerial vehicles such as drones. Prior art literature

[0007] Republic of Korea Published Patent No. 10-2022-0095057 (Feb. 22, 2021) "System and method for monitoring shipyard storage area using drone imagery" The problem to be solved

[0008] The purpose of the method for detecting recyclable resources based on drone imagery using artificial intelligence according to one embodiment of the present invention is to provide a method for detecting recyclable resources based on drone imagery using artificial intelligence, which generates spatial information from a captured image of a drone, detects recyclable resources in the captured image using artificial intelligence, and calculates the location and scale of the recyclable resources based on the spatial information to enable collection.

[0009] The problems of the present invention are not limited to the technical problems mentioned above, and other unmentioned technical problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0011] A method for detecting recyclable resources based on drone images using artificial intelligence according to an embodiment of the present invention is a method for detecting recyclable resources based on drone images using artificial intelligence performed by a server, comprising: receiving a photograph in which a drone captures a predetermined area; performing spatial information processing on the photograph to generate spatial information data; inputting the spatial information data into a pre-trained artificial intelligence model to detect recyclable resources on the spatial information data and outputting a detection area; assigning actual coordinates to the detection area; and reflecting the spatial information data to the detection area to calculate the volume of the recyclable resources.

[0012] Here, the above detection area may further include a cadastral information lookup step of matching the detection area with a cadastral map and looking up cadastral information to determine whether it is a collection target.

[0013] Here, the step of generating the spatial information data may include: a step of extracting feature points of a plurality of the captured photographs and extracting matching points where corresponding feature points are matched; a step of generating a grid based on the ground sample distance of the captured photographs and generating a numerical surface model by assigning the highest height value among the height values ​​of the plurality of matching points included in the grid; and a step of generating an orthophoto by inputting the RGB value of the matching point having the highest height value among the height values ​​of the plurality of matching points included in the grid.

[0014] Here, the step of generating the spatial information data may further include the step of standardizing the orthophoto in response to the requirements of the artificial intelligence model.

[0015] Here, the normalization step divides the orthophoto into multiple standardized images of the same size having overlapping regions, and if the number of pixels of the standardized images is insufficient, virtual empty cells may be formed.

[0016] Here, the detection area is acquired as photographic coordinates, and the step of assigning the actual coordinates may calculate the actual coordinates of the detection area based on the coordinate information of the orthophoto and the ground sample distance. Effects of the invention

[0018] According to embodiments of the present invention, at least the following effects are achieved.

[0019] According to the method for detecting recyclable resources based on drone images using artificial intelligence according to the present invention, by generating spatial information using images captured by a drone and analyzing it through artificial intelligence, recyclable resources that are illegally discarded and not managed can be detected.

[0020] In addition, by utilizing drone footage to detect recyclable resources, it is possible to easily detect recyclable resources abandoned in extensive areas that are difficult for humans to access.

[0021] In addition, by processing spatial information on the drone's captured photos, the scale of recyclable resources detected in the 2D photos can be easily and accurately identified.

[0022] In addition, actual coordinates are assigned to the recycling resources detected by the AI ​​model based on footage captured by a drone, allowing for the accurate identification of the location information of the corresponding recycling resources.

[0023] In addition, by comparing the detection area of ​​recyclable resources detected in the drone's photographs with intellectual information, it is possible to determine whether the resources are unmanaged due to unauthorized disposal or are private property.

[0024] The effects according to the present invention are not limited to those exemplified above, and various other effects are included in this specification. Brief explanation of the drawing

[0026] FIG. 1 is a flowchart of a method for detecting recyclable resources based on drone imagery using artificial intelligence according to an embodiment of the present invention. FIG. 2 is a schematic configuration diagram of a server performing a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention. FIG. 3 is a flowchart of the spatial information generation step of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention. FIGS. 4 and 5 are exemplary orthophotos of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention. FIGS. 6 and 7 are explanatory diagrams of a standardized orthophoto of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention. FIGS. 8 and 9 are explanatory diagrams of the coordinate assignment step of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention. FIG. 10 is an explanatory diagram of the volume calculation step and collection method setting step of a drone image-based recycling resource detection method using artificial intelligence according to one embodiment of the present invention. Specific details for implementing the invention

[0027] The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the detailed description. The effects and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various forms and should be understood to include all modifications, equivalents, and substitutions that fall within the spirit and scope of the present invention.

[0028] Prior to the explanation, the terms used in the detailed description are explained. In the following embodiments, terms such as "first," "second," etc., are used not in a limiting sense, but for the purpose of distinguishing one component from another. Therefore, it is understood that the first component mentioned below may be the second component within the technical scope of the present invention. Furthermore, singular expressions include plural expressions unless the context clearly indicates otherwise. Additionally, terms such as "include" or "have" mean that the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification exist, and do not preclude the possibility that one or more other features or components may be added.

[0029] Additionally, the size of components in the drawings may be exaggerated or reduced for convenience of explanation. For example, the size and thickness of each component shown in the drawings are depicted arbitrarily for convenience of explanation, so the present invention is not necessarily limited to what is illustrated.

[0030] Hereinafter, an embodiment according to the present invention will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, identical or corresponding components are given the same reference numerals, and redundant descriptions thereof are omitted.

[0031] A method for detecting recyclable resources based on drone imagery using artificial intelligence according to one embodiment of the present invention generates spatial information from a captured image of a drone, detects recyclable resources in the captured image using artificial intelligence, and calculates the location and scale of the recyclable resources based on the spatial information to enable collection.

[0032] FIG. 1 is a flowchart of a method for detecting recyclable resources based on drone imagery using artificial intelligence according to an embodiment of the present invention, and FIG. 2 is a schematic diagram of a server for performing a method for detecting recyclable resources based on drone imagery using artificial intelligence according to an embodiment of the present invention.

[0033] A method for detecting recyclable resources based on drone images using artificial intelligence according to one embodiment of the present invention (S1000) includes a step of receiving captured photos (S100), a step of generating spatial information (S200), a step of detecting recyclable resources (S300), a step of assigning coordinates (S400), a step of calculating volume (S500), a step of looking up cadastral information (S600), and a step of setting a collection method (S700).

[0034] A method for detecting recyclable resources based on drone images using artificial intelligence according to one embodiment of the present invention (S1000) is performed by a server. The server (100) may include at least one processor (110) and a memory (120) that stores instructions instructing at least one processor to perform at least one step.

[0035] Here, at least one processor (110) may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. Each of the memory (120) and the storage device (130) may be composed of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory (120) may be composed of at least one of read-only memory (ROM) and random access memory (RAM).

[0036] The server (100) may include a transceiver (130) that performs communication via a wireless network. Additionally, the server (100) may further include an input interface device (140), an output interface device (150), and a storage device (160) or a database, etc. Each component included in the server (100) may be connected by a bus (170) to communicate with one another.

[0037] The step of receiving captured photos (S100) is a step in which a server receives captured photos, which are aerial images taken by a drone, an unmanned aerial vehicle, to detect unmanaged recycling resources. The drone is equipped with a digital camera to take aerial photos. The photos taken by the drone, an unmanned aerial vehicle, are input into the server as multiple individual photos. The drone can take overlapping shots of the shooting area, and when the drone takes continuous shots while moving in a certain direction, parts of the continuously taken images may overlap each other.

[0038] Specifically, video can be captured according to the drone's flight conditions. In this case, the flight altitude can be set to construct video based on the Ground Sample Distance (GSD).

[0039] FIG. 3 is a flowchart of the spatial information generation step of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention.

[0040] The spatial information generation step (S200) is a step of processing spatial information for a received captured photograph. The spatial information generation step (S200) includes a matching point extraction step (S210), a digital surface model generation step (S220), an orthophoto generation step (S230), and an orthophoto normalization step (S240).

[0041] The matching point extraction step (S210) is a step of generating a three-dimensional dense point cloud (point cloud data, point cloud) for spatial information processing.

[0042] First, feature points are extracted from each received captured image. Feature extraction involves identifying characteristic parts within the captured image. These characteristic parts can be areas within the image that have complex shapes, such as the edges of an object or patterns existing on its surface.

[0043] It is desirable that the extracted feature points be extracted from the same location on the object, even if there are changes in the captured photograph, such as rotation, size, or brightness. Accordingly, in this embodiment, feature points can be extracted using the SIFT (Scale Invariant Feature Transform) algorithm. The SIFT algorithm can be divided into a part that finds the location of key points within the captured photograph and a part that generates descriptors for the surrounding area information of each key point.

[0044] The SIFT algorithm can be robust to changes in image size by generating a Gaussian scale space from the captured image and locating feature point positions. Additionally, the SIFT algorithm can be robust to changes in rotation by considering the orientation at each feature point location during the process of generating feature point descriptors.

[0045] That is, in this embodiment, feature points that are invariant to the size and rotation of a captured photograph can be extracted using SIFT.

[0046] The feature point extraction process extracts candidate feature points in a Gaussian scale space and checks the stability of the candidate feature points to correct the location of the stable feature points to a detailed location. Next, the descriptor generation process obtains a directional component through the gradient of the surrounding area centered on the selected feature points, and generates a descriptor by redefining the region of interest based on the obtained directional component.

[0047] Specifically, Scale-space Extrema Detection is performed on the received captured image. Scale-space Extrema Detection derives the scale space of the captured image by applying a Gaussian filter to the input image. Then, by applying the Difference of Gaussian (DoG), boundary and corner features with significant variation in the image can be extracted as candidate feature points (potential keypoints). It is desirable to select candidate pixels by searching for maximums and minimums (extremes) within the scale space generated by the Difference of Gaussian function.

[0048] Afterwards, each candidate feature point is refined and localized into a feature point, and the gradient direction and magnitude around each feature point are calculated and the most prominent direction is assigned as the direction of the corresponding feature point.

[0049] Then, the final feature points of the photo are extracted by generating descriptors based on the magnitude and direction components of the gradient.

[0050] Subsequently, a technician matches identical or similar feature points between the captured images to generate matching points, from which Sparse 3D points are constructed. In other words, a Sparse 3D point is a group of matching points. Sparse 3D points can be stored as data (3D point cloud).

[0051] Here, bundle adjustment can be performed on sparse 3D points to reduce errors and further increase accuracy. Bundle adjustment can be performed by determining external orientation parameters (EO parameters), such as the shooting position and pose information of a 2D image captured in a photograph.

[0052] The 3D point cloud is represented by individual point cloud data containing position information of X, Y, and Z values ​​and RGB information based on the drone's shooting data.

[0053] The numerical surface model generation step (S220) is a step of generating a numerical surface model through a 3D point cloud based on the matching points generated in the matching point extraction step (S210).

[0054] Generally, a Digital Elevation Model (DEM) removes ground structures such as buildings and represents only terrain elevation values ​​for each grid location, whereas a Digital Surface Model (DSM) includes ground structures such as buildings and represents elevation values ​​for each location on the surface.

[0055] In this embodiment, the numerical surface model can be generated by creating a grid (grid) corresponding to the ground sample distance of the captured photo from the extracted 3D point, and selecting a maximum height value (elevation value) in each grid cell.

[0056] Specifically, a grid is formed based on the ground sample distance from the captured photos. In this case, the grid is generated based on the largest ground sample distance value among multiple captured photos. Although the drone is set to have the target ground sample distance—that is, to have identical ground sample distances—errors may still occur. Consequently, discrepancies may arise in the ground sample distance values ​​of multiple captured photos; in such cases, the grid is generated based on the largest ground sample distance value.

[0057] Then, height values ​​(altitude values) are extracted for each grid. To extract ground points, conjugate points and corresponding conjugate points are selected from the overlapping portions of the two captured images. Here, conjugate points and corresponding conjugate points are the pixel coordinates of the respective images that captured the same ground point, and in this embodiment, they become the matching points described above. As described above, matching points include positional information in X, Y, and Z values ​​as well as RGB information; in this case, multiple matching points may exist within the same grid depending on the number of captured images, angles, etc. In this case, the matching point with the highest altitude value among the matching points is assigned to the corresponding grid.

[0058] The numerical surface model of this embodiment is generated by assigning a maximum height value (altitude value) to each grid.

[0059] FIGS. 4 and 5 are example diagrams of orthophotos of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention.

[0060] In the orthophoto generation step (S230), an orthophoto can be generated by inputting the RGB values ​​of each grid for the grid generated as in the above digital surface model generation step (S220).

[0061] That is, in the digital surface model generation step (S220), the digital surface model is generated by assigning the highest elevation value among the multiple matching points included in each grid, and in the orthophoto generation step (S230), the orthophoto is generated by assigning the RGB values ​​of the matching point having the highest elevation value among the multiple matching points included in each grid.

[0062] FIGS. 6 and 7 are explanatory diagrams of standardized orthophotos of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention.

[0063] The orthophoto standardization step (S240) is a step of standardizing the orthophoto according to the size of the object to be detected in the artificial intelligence model described later, in accordance with the inference area requirements.

[0064] Spatial information is sensitive to scale, and since orthophotos have different sizes depending on the drone's shooting range (area of ​​the shooting region), the size of the orthophoto may not always be constant.

[0065] For example, if the ground sample distance is 4 cm / px and the drone’s shooting range is 400 m, the image has a large size with 10,000 pixels in both width and height; if a drone photographs several km to cover an administrative district, the size of the orthophoto becomes very large. In such cases, compressing the orthophoto may be impossible, and even if it is compressed to a smaller size, information from the captured data may be lost during the compression process. Furthermore, if the size of the same object varies across different captured photos during the compression process, the detection capability of the AI ​​model may be degraded. In other words, the accuracy of the detection results may decrease.

[0066] Accordingly, in this embodiment, the orthophoto is standardized to a size in which the AI ​​model can optimally detect the target. The standardization of the orthophoto corresponds to the requirements of the AI ​​model by dividing the orthophoto to achieve the number of pixels that maximizes the AI ​​model's detection capability.

[0067] For example, when a standardized image with a size of 640x640 pixels is standardized to 128x128 pixels, which is the size required by an artificial intelligence model, the standardized image is generated by dividing the standardized image into multiple images. At this time, the standardized images that are adjacent to each other are standardized so that they overlap. In this embodiment, the standardized images can be standardized so that one pixel overlaps in each horizontal and vertical direction of the standardized image.

[0068] Meanwhile, during the process of normalizing an orthophoto, if the number of pixels in the final normalized image is insufficient, the corresponding pixels form a virtual empty cell. For example, the corresponding pixels can be treated as 0.

[0069] In the above example, when generating a 128x128 pixel image from a 640x640 orthophoto, the normalization is performed so that one pixel overlaps in each direction. Consequently, the last normalized image in the horizontal direction is short by 4 pixels in the horizontal direction. The missing pixels form a virtual empty cell, ensuring that all normalized images have the same size.

[0070] This is because if standardized images do not have the same size, the size of objects within the images may change during the size equalization process.

[0071] Standardized orthophotos store actual coordinates and ground sample distance information based on the drone's shooting data.

[0072] And, the normalized orthophotos are dissolved again, and since adjacent normalized images have overlapping regions, they can be dissolved accurately.

[0073] The recycling resource detection step (S300) is a step in which a standardized orthophoto is input into a pre-trained artificial intelligence model, and recycling resources on the orthophoto are detected by the artificial intelligence model.

[0074] In this embodiment, the artificial intelligence model may use an image segmentation method type. Representative examples of artificial intelligence models applied to automatically extract recyclable resources from drone footage include semantic segmentation and instance segmentation.

[0075] In other words, the AI ​​model receives a standardized orthophoto as input, performs segmentation, and detects recyclable resources on the standardized orthophoto.

[0076] An artificial intelligence model can be trained to identify the types of recyclable resources included in input images by inputting images containing various recyclable resources into the input stage and applying information about the types of recyclable resources to the output stage to perform supervised learning.

[0077] In this embodiment, the artificial intelligence model can be generated by performing supervised learning as a segmentation detection type artificial intelligence model using training data in which labels have been completed on images constructed individually according to the type of recycled resource. The segmentation detection type artificial intelligence model may be any one of RCNN, Fast RCNN, Fater RCNN, Mask R-CNN, FCN (Fully Convolutional Network), U-net, and FusionNet.

[0078] The AI ​​model performs segmentation detection on the input captured photos and outputs information regarding recycling resources for each grid of the captured photos.

[0079] The segmentation area of ​​the artificial intelligence model is obtained using photo coordinates.

[0080] FIGS. 8 and 9 are explanatory diagrams of the coordinate assignment step of a drone image-based recycling resource detection method using artificial intelligence according to an embodiment of the present invention.

[0081] The coordinate assignment step (S400) is a step of assigning actual coordinates to the recycling resources detected by the artificial intelligence model.

[0082] The segmentation area detected by the artificial intelligence model (i.e., the recycling resource detection area) is obtained as photo coordinates. Photo coordinates are represented by the number of pixels in the x-direction (horizontal direction) and the y-direction (vertical direction).

[0083] Polygons are extracted for the segmentation regions detected by the AI ​​model. Specifically, polygons are extracted by connecting the points of the grids within the segmentation regions output by the AI ​​model, and are stored as spatial polygon data containing coordinate information.

[0084] A standardized orthophoto is a single orthophoto divided into multiple parts. The AI ​​model performs segmentation detection on each of these standardized orthophotos; however, since the segmentation area is obtained using photographic coordinates, the top-left corner is recognized as (0,0), which differs from the actual coordinates in the orthophoto. Therefore, it is necessary to convert the detection area output in photographic coordinates into actual coordinates.

[0085] As described above, the standardized orthophoto contains actual coordinates. Accordingly, the actual coordinate information of the detection area is calculated based on the actual coordinates of the standardized orthophoto and the ground sample distance. Then, the area is converted into spatial data containing actual coordinate information.

[0086] For example, in a standardized orthophoto, if the reference position is the top right, the top left becomes Xmax and Ymin, and the bottom right becomes Xmin and Ymax. When a detection area is detected by an artificial intelligence model and photo coordinates are obtained, the actual coordinate information of the detection area is calculated. The photo coordinates have the top left as (0,0), and each coordinate of the detection area is represented by the number of pixels in the x-direction (horizontal direction) and the y-direction (vertical direction). Accordingly, the actual coordinates of the detection area are determined as Xmax - (ground sample distance X number of pixels in x-direction) and Ymin + (ground sample distance X number of pixels in y-direction).

[0087] Furthermore, since the standardized orthophotos have overlapping regions, a dissolve is performed as described above, and detection regions that are identical recycling resource detection regions but are divided by the standardization of the orthophotos are merged.

[0088] FIG. 10 is an explanatory diagram of the volume calculation step and collection method setting step of a drone image-based recycling resource detection method using artificial intelligence according to one embodiment of the present invention.

[0089] The volume calculation step (S500) is a step for calculating the volume of recycled resources by calculating the volume of spatial information data.

[0090] The volume of recyclable resources is calculated by analyzing spatial data containing actual coordinate information and a digital surface model. The area is calculated from the actual coordinates of the spatial data, and the volume can be calculated from the height values ​​(elevation values) of the digital surface model.

[0091] And the price of recycled resources is calculated.

[0092] Density and price information corresponding to the types of recyclable resources can be pre-stored on the server. This can be stored in the server's database. Based on this, the price of the recyclable resource can be calculated by multiplying the calculated volume by the previously stored density and price information.

[0093] The cadastral information inquiry step (S600) is a step of matching the spatial information of the detection area with a cadastral map and querying information such as the owner, lot number, and land category of the land. The cadastral information may be stored in advance on a server or obtained by linking with a DB held by the Korea Land Information System (KLIS).

[0094] This is because while recyclable resources detected by AI models within standardized orthophotos may be subject to collection if they are illegally dumped and piled up, they should not be collected if they are piled on private property, as there is a high probability that they belong to the landowner.

[0095] Accordingly, the collection of the detected target is determined based on the land parcel number and land category retrieved after matching with the cadastral map. In this case, even if it is a recyclable resource classified as a collection target, if its volume is less than a preset volume, it may be excluded from collection targets considering collection costs, the possibility of recurrence of illegal dumping in the same location, etc.

[0096] The collection method setting step (S700) is a step for setting the route and scale for the recyclable resources determined as collection targets. The scale of the recyclable resources can be calculated from volume information and density information, and the vehicles, personnel, etc. required for collection can be efficiently determined by checking the collection scale in advance before collection. Additionally, the collection route can be determined according to pre-set rules, and as an example, the route can be set clockwise from the starting position.

[0097] As described above, according to the present invention, a method for detecting recyclable resources based on drone images using artificial intelligence is provided, wherein spatial information is generated from a captured image of a drone, recyclable resources in the captured image are detected using artificial intelligence, and the scale of recyclable resources is calculated based on the spatial information and collected.

[0098] The use of all examples or exemplary terms (e.g., etc.) in the present invention is merely for the purpose of describing the invention in detail, and the scope of the invention is not limited by such examples or exemplary terms unless limited by the claims. Furthermore, a person skilled in the art will understand that various modifications, combinations, and changes may be made according to design conditions and factors within the scope of the claims or equivalents to which they are added.

[0099] Accordingly, the scope of the present invention is not limited to the embodiments described above, and all scopes equivalent to or equivalently modified from the claims set forth below, as well as the claims set forth below, shall be considered to fall within the scope of the concept of the present invention. Explanation of the symbols

[0101] S1000: Drone Image-Based Recycling Resource Detection Method Using Artificial Intelligence S100: Captured photo reception stage S200: Spatial information generation stage S300: Recycling resource detection step S400: Coordinate assignment step S500: Volume calculation step S600: Cadastral information inquiry step S700: Collection method setting step

Claims

Claim 1 A method for detecting recyclable resources based on drone imagery using artificial intelligence performed by a server, comprising: receiving a photograph in which a drone captures a predetermined area; performing spatial information processing on the photograph to generate spatial information data; inputting the spatial information data into a pre-trained artificial intelligence model to detect recyclable resources on the spatial information data and outputting a detection area; assigning actual coordinates to the detection area based on the coordinate information of the orthophoto generated in the step of generating the spatial information data and the Ground Sample Distance (GSD); and reflecting the spatial information data to the detection area to calculate the volume of the recyclable resources; wherein the step of generating the spatial information data includes: extracting feature points using a SIFT (Scale Invariant Feature Transform) algorithm on a plurality of the photographed images, extracting matching points formed by matching corresponding feature points, and then constructing a 3D point cloud which is a group of the matching points. A method for detecting recyclable resources based on drone images using artificial intelligence, comprising: a step of generating a grid based on the ground sample distance of the above-described photograph and generating a numerical surface model by assigning the highest height value among the height values ​​of a plurality of matching points included in the grid; a step of generating the above-described orthophoto by inputting the RGB value of the matching point having the highest height value among the height values ​​of a plurality of matching points included in the grid; and a step of standardizing the above-described orthophoto in accordance with the requirements of the artificial intelligence model; wherein the standardization step divides the above-described orthophoto into a plurality of standardized images of the same size having overlapping regions, and forms a virtual empty cell when the number of pixels of the standardized images is insufficient. Claim 2 A method for detecting recyclable resources based on drone imagery using artificial intelligence, further comprising, in claim 1, a step of matching the detection area with a cadastral map and querying cadastral information to determine whether it is a collection target. Claim 3 A method for detecting recyclable resources based on drone images using artificial intelligence, wherein, in claim 1, the detection area is obtained as photo coordinates, the artificial intelligence model extracts polygons for the detection area and stores them as spatial information polygon data having coordinate information. Claim 4 In claim 1, the step of calculating the volume is an artificial intelligence-based drone image-based recycling resource detection method in which the area is calculated from the actual coordinates of the spatial information data and the volume of the recycling resource is calculated from the height value of the numerical surface model. Claim 5 delete Claim 6 delete

Images