Information processing device, information processing method, and program

Abstract

To more appropriately select a representative point on the ground that is to be the object of examination.SOLUTION: Provided is an information processing device comprising: an image acquisition unit for acquiring multiple types of images obtained by capturing images of the ground surface from above; a classification processing unit for setting vector information of an order corresponding to the type of image, for a pixel or a pixel group in an image, and performing unsupervised classification processing on the pixel or pixel group in the image on the basis of vector information; a smoothing processing unit for performing smoothing processing on a classification result image resulting from the unsupervised classification processing for reducing variation in the classification result in neighboring pixels; and a representative point selection unit for selecting one or more representative points from within regions having the same classification result on the basis of the smoothed image resulting from smoothing processing. Processing for rewriting with the mode value of neighboring pixels or pixel groups is performed on a pixel or pixel group in the classification result image, and the pixel or pixel group with which the frequency of the mode value is smaller than or equal to a first threshold is assigned an error label.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] In agriculture, there is a high need to understand the soil characteristics of fields from the perspective of field management and appropriate fertilization. The Ministry of Agriculture, Forestry and Fisheries' Basic Plan for Food, Agriculture and Rural Areas also promotes "data-based soil creation." In agricultural fields, "soil diagnosis" is widely practiced, in which soil samples are collected from several points within the field, mixed together, and used as a representative soil sample to analyze its physicochemical properties, and many companies offer soil diagnosis services.

[0003] In relation to this, a technology has been disclosed in which visible images of a field immediately after rain and when it is dry are taken using an aerial photography helicopter, soil is sampled at several points within the field, the volumetric moisture content and gas phase content are analyzed, the red, green, and blue color separation value data at the soil sampling points in the image are light intensity corrected to obtain the difference in light intensity corrected color separation value data, and a map of the volumetric moisture content and gas phase content is created to obtain a regression equation, thereby obtaining a dry / wet map for the entire field (Patent Document 1).

[0004] Furthermore, a technology has been disclosed in which the amount of gamma rays emitted from the soil is measured and a gamma ray dose distribution map of the target area is created to grasp the distribution of basic soil composition types derived from bedrock, and the distribution of basic soil composition types is used as an indicator to determine the necessary but efficient collection points and number of soil samples, and the actual content or characteristics of the soil composition for each basic soil composition type are analyzed by analyzing the soil samples collected based on the determination, and the analysis results are applied to a gamma ray dose distribution map, thereby grasping the actual soil composition distribution or creating such a distribution map (Patent Document 2).

[0005] Also, a method has been disclosed for classifying farm field environments through unsupervised learning using high-resolution color infrared (CIR) images, which are aerial images of farm fields (Non-Patent Document 1). [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2012-150068 [Patent Document 2] Japanese Patent Application Laid-Open No. 2004-12160 [Non-patent literature]

[0007] [Non-Patent Document 1] “In-field variability detection and spatial yield modeling for corn using digital aerial imaging”, Gopalapillai and Tian (1999), Transaction of ASAS 42: 1911-1920 Summary of the Invention [Problem to be solved by the invention]

[0008] In soil diagnostic technology, it is not practical to thoroughly survey the entire field, so it is desirable to select survey points. For example, methods such as the diagonal method and the random method are known. However, conventional technologies do not select survey points based on any soil characteristics, so there is little basis for selecting representative survey points, raising concerns that the diagnostic results may not accurately capture the field environment. The technology described in Non-Patent Document 1 classifies field environments, but because it does so using a single type of image, there are concerns that accuracy is insufficient. Furthermore, it does not attempt to select representative points. Furthermore, the technology described in Patent Document 2 processes data based on gamma radiation from bedrock and covers a wide area. Therefore, it is not suitable for analyzing soils on a field-level scale.

[0009] The present invention has been made in consideration of these circumstances, and one of its objects is to provide an information processing device, an information processing method, and a program that can more appropriately select representative points on the ground to be surveyed. [Means for solving the problem]

[0010] An information processing device according to one aspect of the present invention comprises an image acquisition unit that acquires multiple types of images, each of which is obtained by capturing an image of the ground from above; a classification processing unit that sets, for each pixel or pixel group in the image, vector information of an order corresponding to the type of image, and performs unsupervised classification processing on the pixel or pixel group in the image based on the vector information; a smoothing processing unit that performs smoothing processing on a classification result image that is the result of the unsupervised classification processing to reduce variation in classification results among neighboring pixels; and a representative point selection unit that selects one or more representative points from within an area with the same classification result based on the smoothed image that is the result of the smoothing processing, wherein the smoothing processing unit repaints the pixel or pixel group in the classification result image with the most frequent value of neighboring pixels or pixel groups, and assigns an error label to a pixel or pixel group whose frequency of the most frequent value is equal to or less than a first threshold.

[0011] In the above aspect of the present invention, the representative point selection unit may select a point that is a survey target for soil survey as the representative point.

[0012] In the above aspect of the present invention, the classification processing unit may specify the number of categories to perform the unsupervised classification process.

[0013] In the above aspect of the present invention, the image acquisition section may acquire a plurality of types of images including at least a thermal image or a DSM image.

[0014] In the above aspect of the present invention, the method further includes an evaluation unit that evaluates the smoothed image that is the result of the classification processing unit and the smoothing processing unit repeatedly performing processing while changing the type of image and the number of classifications, and the evaluation unit may evaluate the smoothed image as "invalid" if the proportion of pixels or pixel groups assigned with error labels is equal to or greater than a second threshold.

[0015] An information processing method according to another aspect of the present invention includes an information processing device that acquires multiple types of images, each of which is obtained by capturing an image of the ground from above, sets vector information of an order corresponding to the type of image for each pixel or group of pixels in the images, performs an unsupervised classification process on the pixels or groups of pixels in the images based on the vector information, performs a smoothing process on a classification result image that is the result of the unsupervised classification process to reduce variation in classification results among neighboring pixels, selects one or more representative points from within an area with the same classification result based on the smoothed image that is the result of the smoothing process, repaints the pixel or group of pixels in the classification result image with the most frequent value of neighboring pixels or groups of pixels, and assigns an error label to a pixel or group of pixels whose frequency of the most frequent value is equal to or less than a first threshold.

[0016] A program according to another aspect of the present invention causes an information processing device to acquire multiple types of images, each of which is obtained by capturing images of the ground from above, set vector information of an order corresponding to the type of image for pixels or groups of pixels in the images, perform unsupervised classification processing on the pixels or groups of pixels in the images based on the vector information, perform a smoothing process on a classification result image that is a result of the unsupervised classification processing to reduce variation in classification results among neighboring pixels, select one or more representative points from within an area with the same classification result based on the smoothed image that is the result of the smoothing process, perform a process of repainting a pixel or group of pixels in the classification result image with the mode value of neighboring pixels or groups of pixels, and assign an error label to a pixel or group of pixels whose frequency of the mode is equal to or less than a first threshold. [Effects of the Invention]

[0017] According to each of the above aspects, it is possible to more appropriately select representative points on the ground to be surveyed. [Brief explanation of the drawings]

[0018] [Figure 1] 1 is a diagram illustrating an example of a usage environment and configuration of an information processing device 100. FIG. [Figure 2] FIG. 10 is a diagram showing an example of a classification result image obtained as a result of performing classification processing with k=5 specified. [Figure 3] FIG. 10 is a diagram illustrating an example of a smoothed image. [Figure 4] FIG. 2 is a diagram showing an example of representative points selected by a representative point selecting section 150. [Figure 5] FIG. 10 is a diagram showing trends in soil properties for each image classification result. [Figure 6] FIG. 10 is a diagram showing an example of a usage environment and configuration of an information processing device 100A according to a second embodiment. [Figure 7] FIG. 10 is a diagram showing an example of a smoothed image to which an error label has been added by the smoothing processing unit 130 of the second embodiment. [Figure 8] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 9] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 10] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 11] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 12]This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 13] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 14] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. [Figure 15] This figure shows smoothed images generated while changing the order and number of classes, and the analytical values ​​of soil samples obtained from the distribution areas of each class. DETAILED DESCRIPTION OF THE INVENTION

[0019] Hereinafter, an information processing device, an information processing method, and a program according to an embodiment of the present invention will be described with reference to the drawings.

[0020] First Embodiment 1 is a diagram showing an example of the usage environment and configuration of an information processing device 100. The information processing device 100 acquires, for example, an image of the ground (e.g., farmland) F captured from the sky (aerial photography) from an aircraft 10 via a relay device 20, and provides the information to a terminal device 30. Note that the method of acquiring the image is not limited to this, and the image may be acquired by loading an image stored on a storage medium into a drive device of the information processing device 100. Furthermore, the image may have been preprocessed before being passed to the information processing device 100.

[0021] The aircraft 10 is a drone (unmanned aerial vehicle) equipped with various imaging devices. Imaging devices include visible light cameras, ranging cameras, multispectral cameras, and thermal infrared cameras. The aircraft 10 may be equipped with all of these cameras, or multiple aircraft 10 may each be equipped with a different imaging device and take aerial photographs in sequence. The aircraft 10 stores multiple aerial images captured by the imaging devices 12, along with the camera parameters of the aerial images and location information such as GPS, on a memory card installed in the aircraft 10 or transmits the images to the relay device 20.

[0022] The relay device 20 is a terminal device such as a tablet terminal that operates the flying vehicle 10 and has an application installed for receiving and displaying aerial images transmitted from the flying vehicle 10. The relay device 20 displays the aerial images as RGB images and transmits the received aerial images to the information processing device 100 via the network NW. The network NW is any network such as a LAN, a WAN, or an internet line, and may be wired or wireless.

[0023] The terminal device 30 is, for example, a computer device such as a personal computer, a smartphone, or a tablet terminal. The terminal device 30 communicates with the information processing device 100 via a network NW and displays information received from an interface unit 160 of the information processing device 100, which will be described later. The terminal device 30 may be attached to the information processing device 100.

[0024] The information processing device 100 has, for example, a function of a web server. The information processing device 100 acquires multiple types of images from the relay device 20 or the like, performs classification processing, smoothing processing, representative point selection processing, and the like using methods described later, and outputs the processing results to the terminal device 30.

[0025] The information processing device 100 includes, for example, an image acquisition unit 110, a classification processing unit 120, a smoothing processing unit 130, a representative point selection unit 150, and an interface unit 160. These components are realized by, for example, a hardware processor such as a CPU (Central Processing Unit) executing a program (software). Some or all of these components may be realized by hardware (including circuitry) such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a GPU (Graphics Processing Unit), or may be realized by a combination of software and hardware. The program may be stored in advance in a storage device (a storage device having a non-transitory storage medium) such as an HDD (Hard Disk Drive) or flash memory, or may be stored in a removable storage medium (a non-transitory storage medium) such as a DVD or CD-ROM, and installed in the storage device by inserting the storage medium into a drive device.

[0026] The image acquisition unit 110 acquires multiple types of images, each of which is obtained by capturing images of farmland from above. The multiple types of images include some or all of visible light images, DSM (Digital Surface Model), images for each frequency band taken by a multispectral camera, thermal images, and near-infrared images. DSM is an example of a height image generated by Sfm (Structure from Motion). In the following description, it is assumed that six types of images are acquired: a red edge image, a red image, a near-infrared image, a green image, DSM, and a thermal image. A red edge image is an image that captures the reflectance of a wavelength slightly longer than the wavelength of a typical red (for example, around 735 [nm]). A red image is an image that captures the reflectance of a wavelength belonging to red. A green image is an image that captures the reflectance of a wavelength belonging to green. A near-infrared image is an image that captures the reflectance of near-infrared light (around 790 [nm]). Preprocessing such as Sfm and edge extraction may be performed by the information processing device 100 or by another device. Furthermore, it is preferable that the various images are orthoimages made by joining together images from multiple frames, because if an image from one frame is used, there is a concern that the accuracy may decrease due to image distortion.

[0027] The classification processing unit 120 sets vector information (n-dimensional vector) of an order corresponding to the type of image for pixels or pixel groups in the various images, and performs unsupervised classification processing for the pixels or pixel groups in the images based on the n-dimensional vector. Here, since six types of images are to be processed, n=6. "Regarding pixels or pixel groups" means that the processing target may be a pixel in an image, or several groups of pixels that are adjacent to each other. In the following explanation, it is assumed that the processing target is a pixel. The classification processing unit 120 generates a six-dimensional vector using pixel values, altitude values, etc. in the various images as element values ​​of the vector. At this time, normalization processing may be performed on the element values.

[0028] The classification processing unit 120 classifies each pixel into one of classes 1 to k, without considering the pixel's position, using, for example, the k-means++ method, which is performed by specifying the number of classes k. Figure 2 shows an example of a classification result image that is the result of classification processing performed by specifying k=5. In the figure, f indicates the field of interest within farmland F. The classification result image contains many high-frequency fluctuation components, and it is highly likely that it is not appropriate to classify farmlands with similar characteristics based on this alone.

[0029] The smoothing processing unit 130 performs a smoothing process on the classification result image to reduce variations in classification results among neighboring pixels. For example, the smoothing processing unit 130 performs a process called a majority filter. That is, the smoothing processing unit 130 repaints the classification result image with the most frequent value of neighboring pixels (for example, pixels within a radius of 30 pixels). The smoothing processing unit 130 may obtain the smoothed image by requesting the smoothing process from a device separate from the information processing device 100. FIG. 3 is a diagram showing an example of a smoothed image that is the result of the smoothing process. As shown in the figure, the smoothed image shows that points in the field f with similar characteristics are grouped into roughly similar regions.

[0030] Based on the smoothed image, the representative point selection unit 150 selects one or more representative points from each area with the same classification result (from areas belonging to the same class). Hereinafter, areas belonging to the same class will be referred to as class areas. A representative point is a point on the image that represents a location to be surveyed for soil survey. For example, for a certain class area, the representative point selection unit 150 selects as the representative point the point that is farthest from the boundary line with other adjacent class areas. Figure 4 is a diagram showing an example of representative points selected by the representative point selection unit 150.

[0031] The interface unit 160 provides information on the representative points selected by the representative point selection unit 150 together with, for example, actual location information (latitude, longitude) to the terminal device 30. After viewing this information, the user can determine the amount of fertilizer to be applied for each class area or the operation schedule of the cultivation equipment.

[0032] The effectiveness of the present invention will now be described. The inventors of the present invention conducted soil surveys and analyses at several dozen points in a certain field f and compared soil properties based on the classification result images displayed after processing in accordance with the first embodiment. Figure 5 shows the trends in soil property values ​​for each class and the entire field. Class 5 is not shown because its class range is small. The upper left graph in Figure 5 represents various index values: total carbon content [g / kg], the upper right graph represents total nitrogen content [g / kg], the middle left graph represents clay content [%], the middle right graph represents electrical conductivity [mS / cm], the lower left graph represents moisture content [%], and the lower right graph represents pH (H2O). The horizontal axis of each graph represents classes 1 to 4 or the entire field f. In the boxplots, O1 represents the interquartile range and median, O2 represents the range from maximum to minimum, and O3 represents the plotted points of the analyzed soil properties for each class. As shown in the figure, there is a significant tendency between classes for all index values, and it is clear that the classification process is appropriate.

[0033] According to the first embodiment described above, the system is equipped with an image acquisition unit 110 that acquires multiple types of images, each of which is obtained by photographing farmland from above, a classification processing unit 120 that sets vector information of order n according to the type of image for each pixel or group of pixels in the image and performs unsupervised classification processing on the pixels or group of pixels in the image based on the vector information, a smoothing processing unit 130 that performs smoothing processing on the classification result image that is the result of the unsupervised classification processing to reduce variation in classification results among neighboring pixels, and a representative point selection unit 150 that selects one or more representative points from within an area with the same classification result based on the smoothed image that is the result of the smoothing processing, thereby making it possible to more appropriately select representative points to be surveyed on the ground (e.g., farmland, a field).

[0034] Second Embodiment The second embodiment will be described below. In the first embodiment, the order n and the number of classes k are assumed to be fixed values, but in the second embodiment, the classification processing unit 120 and the smoothing processing unit 130 repeatedly perform processing while changing the order (type of image) n and the number of classes (number of classifications) k, and an evaluation unit 140 is further provided that evaluates the resulting smoothed image. Figure 6 is a diagram showing an example of the usage environment and configuration of an information processing device 100A according to the second embodiment.

[0035] In the second embodiment, the classification processing unit 120 selects any two or more of the multiple types of images described above (for example, red edge image, red image, near-infrared image, green image, DSM, and thermal image), and sets an n-dimensional vector with the selected number set to n. The classification processing unit 120 also sets the number of classes k to any value between 3 and 5, for example. In this way, the classification processing unit 120 and the smoothing processing unit 130 repeatedly perform the above processing while changing the type of n-dimensional vector and the number of classes k, and generate multiple smoothed images.

[0036] The evaluation unit 140 evaluates each of the multiple smoothed images. For example, in the process described in the first embodiment, the smoothing processing unit 130 of the second embodiment assigns an error label to pixels whose frequency of the most frequent value is equal to or less than a first threshold value Th1. If the ratio of pixels assigned with an error label to the total pixels of the smoothed image is equal to or greater than a second threshold value Th2, the evaluation unit 140 evaluates the smoothed image as "invalid." FIG. 7 is a diagram showing an example of a smoothed image to which an error label has been assigned by the smoothing processing unit 130 of the second embodiment. A high ratio of pixels assigned with an error label indicates that there is a tendency for pixels of various classes to exist in the neighboring area of ​​the pixel, which is a phenomenon that suggests that the classification result of the classification processing unit 120 is invalid.

[0037] The interface unit 160 of the second embodiment, for example, excludes smoothed images evaluated as "invalid" by the evaluation unit 140 and displays them on the terminal device 30. In this way, the interface unit 160 presents smoothed images to the user. The interface unit 160 then accepts a selection operation by the user to select one or more smoothed images and passes the result of the selection operation to the representative point selection unit 150. The representative point selection unit 150 selects representative points based on the smoothed images passed from the interface unit 160 and outputs the selected representative points to the terminal device 30.

[0038] In the second embodiment, the evaluation unit 140 may be omitted, and all smoothed images may be presented to the user to select one. Alternatively, instead of having the user select a smoothed image, one or more smoothed images may be automatically selected based on some index from the remaining smoothed images excluding those evaluated as "invalid," and the representative point selection unit 150 may select a representative point based on the selected smoothed image.

[0039] The inventors of the present invention conducted an actual soil survey and measured soil properties at points corresponding to representative points selected by processing in accordance with the second embodiment for a certain field f. Figures 8 to 15 show smoothed images generated by varying the order (image type) n and the number of classes (number of classifications) k, as well as analytical values ​​of soil samples obtained from the distribution areas of each class. The analytical values ​​of the soil samples show various index values, such as moisture content (%) (top left), coarse sand fraction (%) (top middle), fine sand fraction (%) (top right), silt fraction (%) (middle left), clay fraction (middle middle), pH (HO) (middle right), electrical conductivity (mS / cm) (bottom left), total carbon (g / kg) (bottom middle), and total nitrogen (lower right).

[0040] Figure 8 shows smoothed images generated with five classes for DSM and thermal images, along with analytical values ​​for soil samples obtained from the distribution areas of each class. Figure 9 shows smoothed images generated with five classes for near-infrared images, DSM, and thermal images, along with analytical values ​​for soil samples obtained from the distribution areas of each class. Figure 10 shows smoothed images generated with three classes for four visible light images of different frequency bands, along with analytical values ​​for soil samples obtained from the distribution areas of each class. Figure 11 shows smoothed images generated with four classes for four visible light images of different frequency bands, along with analytical values ​​for soil samples obtained from the distribution areas of each class. Figure 12 shows smoothed images generated with five classes for four visible light images of different frequency bands, along with analytical values ​​for soil samples obtained from the distribution areas of each class. 13 to 15 show smoothed images generated for the same six types of images as in the first embodiment, with the number of classes being 3 to 5, and the analytical values ​​of soil samples obtained from the distribution areas of each class.

[0041] In this way, the distribution of class regions in a smoothed image differs depending on the type of image and the number of classes. According to the second embodiment, by presenting a smoothed image to the user and allowing them to select a representative point that matches the user's image of the soil distribution in the field, it is possible to select a representative point that matches the user's image of the soil distribution in the field.

[0042] Although the above description has been focused on determining representative points for soil surveys, the present invention can also be applied to other applications. For example, the present invention can be applied to selecting representative points, which are points for surveying the growth of crops, based on multiple types of images taken from the air of a field where crops are planted. For different applications, other types of images may be used in addition to the types of images listed above.

[0043] According to each of the embodiments described above, it is possible to more appropriately select representative points to be surveyed on the ground.

[0044] The above describes the form for carrying out the present invention using an embodiment, but the present invention is not limited to such an embodiment, and various modifications and substitutions can be made within the scope that does not deviate from the gist of the present invention. [Explanation of symbols]

[0045] 100 Information processing device 110 Image acquisition unit 120 Classification processing unit 130 Smoothing processing section 140 Evaluation Department 150 Representative point selection section 160 Interface section 200 Display device

Claims

1. an image acquisition unit that acquires a plurality of types of images each obtained by capturing an image of the ground from above; a classification processing unit that sets vector information of an order corresponding to the type of the image for a pixel or a pixel group in the image, and performs unsupervised classification processing for the pixel or the pixel group in the image based on the vector information; a smoothing processing unit that performs a smoothing process on the classification result image that is a result of the unsupervised classification process to reduce variations in classification results among neighboring pixels; a representative point selection unit that selects one or more representative points from within the area having the same classification result based on a smoothed image that is a result of the smoothing process, the smoothing processing unit performs a process of repainting a pixel or pixel group in the classification result image with a most frequent value of a neighboring pixel or pixel group, and assigns an error label to a pixel or pixel group whose frequency of the most frequent value is equal to or less than a first threshold value; Information processing device.

2. The representative point selection unit selects a point to be surveyed for soil survey as the representative point.

2. The information processing device according to claim 1.

3. the classification processing unit performs the unsupervised classification process by specifying the number of classifications; 3. The information processing device according to claim 1.

4. The image acquisition unit acquires a plurality of types of images including at least a thermal image or a DSM image. The information processing device according to claim 1 .

5. an evaluation unit that evaluates the smoothed image that is a result of the classification processing unit and the smoothing processing unit repeatedly performing the processes while changing the type of image and the number of classifications; the evaluation unit evaluates the smoothed image as “invalid” when the ratio of the pixels or pixel groups assigned with the error label is equal to or greater than a second threshold value; The information processing device according to claim 1 .

6. The information processing device Each captures multiple types of images of the ground from above, For a pixel or a group of pixels in the image, vector information of an order corresponding to the type of the image is set; performing an unsupervised classification process on pixels or groups of pixels in the image based on the vector information; performing a smoothing process on the classification result image resulting from the unsupervised classification process to reduce variations in classification results among neighboring pixels; selecting one or more representative points from within the area having the same classification result based on a smoothed image resulting from the smoothing process; a process of repainting a pixel or a group of pixels in the classification result image with the most frequent value of a neighboring pixel or group of pixels; assigning an error label to a pixel or a group of pixels whose frequency of the mode is equal to or less than a first threshold; Information processing methods.

7. In the information processing device, Each of these captures multiple types of images of the ground from above, For a pixel or a group of pixels in the image, vector information of an order corresponding to the type of the image is set; performing an unsupervised classification process on a pixel or group of pixels in the image based on the vector information; performing a smoothing process on the classification result image resulting from the unsupervised classification process to reduce variations in classification results among neighboring pixels; selecting one or more representative points from within the area having the same classification result based on a smoothed image resulting from the smoothing process; performing a process of repainting a pixel or a group of pixels in the classification result image with the most frequent value of a neighboring pixel or group of pixels; assigning an error label to a pixel or a group of pixels whose frequency of the mode is equal to or less than a first threshold; program.

Images