JP2022176681A - Acquisition method and acquisition system of image data - Google Patents

Abstract

To provide an acquisition method and an acquisition system of image data which can stably acquire image data that does not reduce estimation accuracy by an estimation model when acquiring the image data used for the estimation model generated by using image data processing.SOLUTION: In each imaging step in which image data used for generation and use of an estimation model is imaged and acquired by an imaging device 2, an object G and at least an image acquisition part 2a of an imaging device 2 are arranged in a storage part 3 having a light shielding property, the illuminance to the object G is set to be same and the imaging background specification is set to be the same by using illumination means 4 that illuminates the inside of the storage part 3 in a state where the external light of the storage part 3 is shielded, each image data that is acquired is stored in a calculation part 7.SELECTED DRAWING: Figure 2

Description

TECHNICAL FIELD The present invention relates to a method and system for acquiring image data used in an estimation model generated using image data processing, and more particularly, to stably acquiring image data without degrading the estimation accuracy of the estimation model. The present invention relates to an acquisition method and an acquisition system for obtainable image data.

In recent years, AI technology has developed remarkably, and various desired estimations are performed using estimation models generated using various machine learning techniques such as deep learning. Some estimation models, which are a type of computer program, are generated using a large number of image data as learning data (teaching data) (see Patent Documents 1 and 2, for example).

Image data is acquired by photographing using a photographing device such as a digital camera. In a well-equipped indoor environment, it is easy to obtain image data under constant imaging conditions. Outdoors, on the other hand, it is difficult to maintain constant shooting conditions due to changes in the intensity and direction of the outside light, the presence of surrounding objects, and other factors. (noise) may be included. The disturbance factor included in the image data is not the information originally intended by the image data. Therefore, if estimation is performed using an estimation model generated using image data that includes disturbance factors and image data that does not include disturbance factors, the estimation accuracy will decrease due to the presence or absence of these disturbance factors. there is a risk of

In addition, when performing estimation by inputting image data to be estimated into an estimation model generated using image data, image data to be estimated (that is, image data used when using the estimation model) contains a disturbance factor, there is a risk that the estimation accuracy will decrease due to the presence or absence of this disturbance factor. When the image data used for generating the estimation model and the image data used when using the estimation model are acquired at different locations, for example, one image data is acquired indoors and the other image data is acquired outdoors. In some cases, the shooting conditions at each location may deviate greatly. Along with this, there is a greater risk of deterioration in estimation accuracy due to disturbance factors contained in image data. Therefore, there is room for improvement in stably acquiring image data without lowering the estimation accuracy of the estimation model.

JP 2013-242825 A JP 2020-85856 A

An object of the present invention is to provide image data that can stably obtain image data that does not reduce the estimation accuracy of an estimation model when obtaining image data used in an estimation model that is generated using image data processing. It is to provide an acquisition method and an acquisition system.

In order to achieve the above object, the image data acquisition method of the present invention provides a method for acquiring image data used for generating and using an estimation model by capturing an object of each image data in each capturing step using a capturing device. and at least the image acquisition unit of the photographing device is arranged inside a light-shielding housing section to block light from the outside of the housing section, and the lighting means is used to illuminate the interior of the housing section. The illuminance for the object is set to be the same, and the shooting background specifications are set to be the same.

An image data acquisition system according to the present invention includes an imaging device that captures and acquires image data used for generating and using an estimation model, a housing section that has a light shielding property, lighting means that illuminates the inside of the housing section, a computing unit that stores the image data and performs data processing using the image data; In a state in which an object and at least an image acquisition section of the photographing device are arranged inside the housing section to block light from the outside of the housing section, the illuminance for the object is set to be the same by the illumination means. , and the shooting background specifications are set to be the same.

According to the present invention, in each photographing step of acquiring image data used for generating and using an estimation model, the object of each image data and at least the image acquisition unit of the photographing device are housed in a light-shielding manner. It is placed inside the housing so as to block light from the outside of the housing. As a result, the inside of the accommodation section where the object is placed becomes a photographing condition in which the influence of external light can be substantially ignored, and the risk of inclusion of disturbance factors in each image data is greatly reduced. By setting the illuminance for the object to be the same using the illumination means for illuminating the inside of the housing portion and by setting the shooting background specifications to be the same, even if each image data contains a disturbance factor, Variation in disturbance factors is eliminated or extremely reduced. As a result, it is advantageous for stably acquiring image data without lowering the estimation accuracy of the estimation model.

1 is an explanatory diagram illustrating an embodiment of an image data acquisition system in a state in which a container is opened in a side view through which the inside of the container is seen through; FIG. FIG. 3 is an explanatory diagram illustrating a state in which the accommodating portion of FIG. 1 is opened; FIG. 3 is an explanatory diagram illustrating the inside of the accommodating portion of FIG. 2 in a plan view; 3 is an explanatory view showing a modification of the internal structure of the accommodating portion of FIG. 2; FIG. FIG. 4 is an explanatory diagram schematically illustrating a neural network used when generating an estimation model; FIG. FIG. 10 is an explanatory diagram illustrating another embodiment of the image data acquisition system in a side view through the inside of the storage unit; FIG. 7 is an explanatory diagram illustrating the inside of the accommodating portion of FIG. 6 in a plan view; FIG. 11 is an explanatory diagram illustrating another embodiment of the image data acquisition system in a side view through the inside of the storage unit; FIG. 9 is an explanatory diagram illustrating the inside of the accommodating portion of FIG. 8 in a plan view;

The image data acquisition method and acquisition system of the present invention will be described below based on the embodiments shown in the drawings.

Image data D used when generating an estimated model M and when using an estimated model M by an image data acquisition system 1 (hereinafter referred to as an acquisition system 1) of the present invention illustrated in FIGS. is obtained. Each image data D is monochrome, grayscale, or color image data in which the object G is photographed. The estimated model M is a type of computer program generated using image data processing. Various machine learning techniques to which AI technology is applied can be used for this image data processing. In this embodiment, the mixed liquid (liquid) is the object G, so the object G is housed in the container 11 . The object G may be solid as well as liquid. In the case of a solid object G, the container 11 can be dispensed with. Details of the object G will be described later.

This acquisition system 1 includes a photographing device 2 that captures and acquires image data D, a housing portion 3 that has a light shielding property, illumination means 4 that illuminates the inside of the housing portion 3, and an arithmetic unit that stores the image data D. 7. A monitor 8 is communicatively connected to the computing unit 7 . In this embodiment the acquisition system 1 further comprises a turntable 5 and a fixed base 10 arranged inside the housing 3 . Rotating table 5 and fixed table 10 can optionally be provided.

The photographing device 2 is means for photographing and obtaining image data D of the object G, and can be camera equipment such as a digital camera, video equipment, or various known means having functions similar to these equipment. . Although one imaging device 2 is provided in this embodiment, a plurality of imaging devices 2 may be provided.

The fixed base 10 is installed at a predetermined position inside the housing portion 3 . The photographing device 2 is placed and fixed on this fixed base 10 . Without the fixed base 10 , the photographing device 2 is placed on the bottom surface of the housing section 3 .

The housing portion 3 blocks light from the outside of the housing portion 3 and forms an internal space for arranging the object G therein. The material of the housing portion 3 may be any material that can block external light of the housing portion 3, and may be resin, metal, wood, paper, fiber, or a combination thereof. As for the light shielding property of the housing portion 3, it is preferable that the light shielding rate measured by the A method of the light shielding property test method specified in JIS L 1055 is 99.9% or more. Although the accommodating part 3 of this embodiment has a rectangular parallelepiped shape, it is not limited to this, and various other shapes can be employed. For example, various shapes such as a cubic shape, a cylindrical shape, and a conical shape can be used, and the accommodating portion 3 can also be formed by covering a light-shielding sheet on a frame that is erected.

The housing portion 3 is composed of a main body 3a and an opening/closing lid 3b. As shown in FIG. . In the photographing step of photographing and acquiring the image data D, the opening/closing lid 3b is closed as shown in FIG.

In this embodiment, the entire photographing device 2 is arranged inside the housing portion 3, but at least the image acquisition portion 2a of the photographing device 2 that captures the image of the object G is arranged inside the housing portion 3. All you have to do is The image acquisition unit 2a is a portion corresponding to a lens in a camera. When only the image acquisition unit 2a of the photographing device 2 is arranged inside the accommodation unit 3, a gap is formed between the image acquisition unit 2a and the accommodation unit 3 so that external light does not enter the inside of the accommodation unit 3. to seal.

The lighting means 4 has a light-emitting portion 4 a that illuminates the inside of the housing portion 3 . The illuminating means 4 may be any means as long as the illuminance inside the housing portion 3 can be set constant. As the light emitting unit 4a, various known electric lights such as an incandescent lamp (light bulb), a fluorescent lamp, and an LED can be used. The number of light emitting units 4a is not limited to one, and a plurality of units can be installed. The light-emitting part 4a can be installed at any position other than the opening/closing lid 3b. preferably.

In this embodiment, the illuminating means 4 further has a diffusion plate 4b, an illuminance meter 4c and a controller 4d, but these can be provided arbitrarily. The diffusion plate 4b is arranged between the light emitting part 4a and the object G, and is detachably attached to the opening/closing lid 3b in this embodiment. The diffuser plate 4b has a function of making the light-emitting portion 4a into a surface light source, and widely diffuses the light emitted by the light-emitting portion 4a to illuminate the object G evenly and more uniformly. A known diffuser film or the like can be used as the diffuser plate 4b.

In this lighting means 4, the illuminance at the position where the illuminance meter 4c is installed is detected by the illuminance meter 4c, and the illuminance can be adjusted by the controller 4d. For example, a target illuminance is set in advance, and the controller 4d is operated based on the illuminance detected by the illuminance meter 4c to control the illuminance for the object G (illuminance on the surface of the object G) to the target illuminance. can also Alternatively, the operator can confirm the illuminance detected by the illuminance meter 4c and manually operate the controller 4d to set the illuminance on the surface of the object G to the target illuminance.

In order to more accurately grasp the illuminance on the surface of the object G, the illuminance meter 4c should be placed as close to the object G as possible, but the illuminance meter 4c may appear in the image data D and become a factor of disturbance. have a nature. Even if the illuminometer 4c is placed at a position slightly away from the object G, the illuminance on the surface of the object G can be relatively easily estimated based on the illuminance detected by the illuminometer 4c. It is preferable to place it in a position close to the object G in an area that does not cause disturbance of the data D. FIG.

The rotating table 5 is installed at a position spaced apart from the fixed table 10 . The turntable 5 is rotated at a desired rotational speed by a drive motor or the like. By placing the object G on the turntable 5 and rotating it, the object G is also rotated. Since the photographing device 2 (image acquisition unit 2a) is arranged to the side of the turntable 5 with a space therebetween, the rotation of the turntable 5 causes the object G to move toward the photographing device 2 (image acquisition unit 2a). Move in the circumferential direction of rotation. If the turntable 5 is not provided, the object G is placed on the bottom surface of the housing portion 3 .

Inside the housing unit 3, the photographing device 2 (image acquisition unit 2a), the lighting means 4 (light emitting unit 4a and diffusion plate 4b), and the turntable 5 are not only fixed at predetermined positions, but can also be installed at desired positions. You can also make it moveable like this: Thereby, the mutual relative position can be adjusted. For example, as shown in FIG. 4, the light emitting unit 4a and the diffuser plate 4b or only the light emitting unit 4a are moved toward and away from the object G placed on the turntable 5 so that the light source is reflected in the image data D. set to a position where there is no (minimum)

Various data such as the image data D and the estimated model M are stored in the calculation unit 7 . The imaging device 2 and the computing unit 7 are connected for wired or wireless communication, and the image data D is input from the imaging device 2 to the computing unit 7 . Alternatively, by removing the memory card of the photographing device 2 and connecting it to the interface of the calculation unit 7, the image data D stored in this memory card may be input from the photographing device 2 to the calculation unit 7. . Therefore, the calculation unit 7 can be arranged not only in the vicinity of the photographing device 2 but also in a place away from the photographing device 2 .

The computation unit 7 performs data processing (computation processing) using various data such as input and stored image data D. FIG. A computer (computer server) is used as the calculation unit 7 . Various data including image data D are stored in a memory of a computer (computer server), and various data processing is performed by a CPU. Various data that have been input, stored, and data-processed in the computing section 7 can be displayed on the monitor 8 . The computing unit 7 can be composed of one computer, but it can also be configured by connecting one or more computers and one or more computer servers in a wired or wireless manner. good. For example, the computing unit 7 is composed of a computer arranged at the construction site together with the photographing device 2 and a computer server arranged at a remote location (laboratory, etc.) from the construction site. A large amount of data may be stored on a computer server and the main data processing may be performed using the computer server.

An estimation model M according to the purpose is generated by various known machine learning techniques. For example, when generating the estimation model M by deep learning using a neural network, as illustrated in FIG. , and weights w (w1, w2, . . . ) are set and connected between nodes 15 in each layer. Then, the image data D (parameters) of the object G as learning data are sequentially input from the input layer 12, pass through the intermediate layer 13, and the estimated values of the items related to the object G (estimated values of the parameters indicating the desired characteristics) to the output layer 14 . The output from each node 15 is transformed using a known activation function (eg, sigmoid function, ReLU, etc.). The output to the output layer 14 is transformed using a known activation function (such as the Softmax function, for example). The estimated value output to the output layer 14 is compared with the actual measurement value corresponding to the estimated value (the actual measurement value of the parameter indicating the desired characteristic), and a known error backpropagation method is used to reduce the error between the two. is used to change the weight w set between the nodes 15 . By performing this arithmetic processing by inputting a large number of image data D, an estimation model M in which the error between the two is reduced, that is, an estimation model M with improved estimation accuracy is generated. The generated estimation model M is stored in the calculation unit 7 .

When estimating using the generated estimation model M, the image data D of the target object G to be estimated is input to the estimation model M. As shown in FIG. The calculation unit 7 performs calculation processing using the input image data D and the estimation model M, and calculates an estimated value of the item related to the object G in the input image data D. FIG. The calculated estimated value is the result of estimation by the estimation model M. FIG. By comparing this estimation result (estimated value) with the measured value corresponding to this estimated value that is separately grasped, the estimation accuracy of the estimation model M can be found.

More specifically, a procedure for acquiring image data D using the present invention when generating and using estimation model M will be described.

The particle size distribution (particle size accumulation curve) of soil can be estimated using an estimation model M generated using a large number of image data D of soil. More specifically, when a liquid mixture of earth and sand mixed with a liquid such as water is placed in a container 11 as an object G and shaken, the container 11 is placed thereon. (For example, cohesive soil) is in a state of adhering to the peripheral surface of the container 11 or in a state of being mixed with the liquid and floating. On the other hand, particles with relatively large particle diameters (for example, sand and gravel) in the earth and sand settle down and accumulate on the bottom of the container 11 . The amount of particles adhering to the peripheral surface of the container 11, the amount of particles floating in the liquid, and the amount of particles settling and depositing on the bottom of the container 11 change according to the particle size distribution of the soil. Therefore, there is a high correlation between the state of the mixed liquid G in the container 11 after the vibration has been applied for a predetermined time and the particle size distribution of the soil. That is, when the particle size distribution of the soil differs, the appearance of the liquid mixture G inside the container 11 (color distribution, shade distribution, adhesion/floating/sedimentation rate) also differs.

First, a large number of image data D to be used as learning data when generating the estimation model M are acquired. Therefore, as exemplified in FIG. 1, a container 11 containing a mixed liquid G in which earth and sand are mixed with a liquid at a prescribed mixing ratio is placed on the turntable 5 after being vibrated under the same uniform conditions. The photographing device 2 is mounted on a fixed base 10 and installed at a predetermined position. The container 11 may be made of transparent glass or synthetic resin so that the internal state can be checked, so it may be a general glass container or a plastic container.

Next, as illustrated in FIG. 2, the liquid mixture G and the photographing device 2 are placed inside the housing portion 3 so that light outside the housing portion 3 is blocked. Inside the storage unit 3 in a state where external light is blocked, the state of the mixed liquid G in the container 11 after a specified time has elapsed after being vibrated is obtained as image data D. In this embodiment, the image data D of each liquid mixture G is acquired by photographing from the side of the container 11 . In the photographing step of acquiring the image data D of each liquid mixture G, the lighting means 4 is used to set substantially the same illuminance for the liquid mixture G, and substantially the same photographing background specifications are set. The photographing background specification means the color (background color) of the background surface of the mixed liquid G when photographing the mixed liquid G, the surface roughness of the background surface, the separation distance between the mixed liquid G and the background surface, and the like. In this embodiment, the background surface is the inner surface of the container 3 located on the back side of the mixed liquid G. As shown in FIG. In each photographing process, the acquisition system 1 is used, the illumination means 4 is operated under the same conditions, and the photographing device 2 and the object G are arranged in the same manner, so that the illuminance for the mixed liquid G is substantially the same. , and the shooting background specifications are also substantially the same.

The particle size distribution (measured value) of the actual soil of the mixed liquid G may be determined by a known method, for example, measured according to the soil particle size test method specified in JIS A 1204. In this way, since the actual measurement result (actual value) corresponding to the estimation of the estimation model M can be grasped, the image data D of the mixed liquid G of various soils with different particle size distributions and the actual measurement value of the particle size distribution of each soil. An estimated model M can be generated by By increasing the number of image data D to be learning data, it is advantageous to improve the estimation accuracy of the estimation model M. FIG.

In this embodiment, in the step of photographing each mixed liquid G when generating the estimated model M, a plurality of image data D are photographed while the mixed liquid G is rotated by the turntable 5 . As a result, it is possible to acquire a plurality of image data D at different positions in the circumferential direction for each mixed liquid G, so that more image data D to be used as learning data can be efficiently acquired. Accordingly, it is advantageous to improve the estimation accuracy of the estimation model M.

When estimating the particle size distribution of the sediment of the mixed liquid G to be estimated using the estimation model M, the sediment was mixed with the liquid at a prescribed mixing ratio in the same manner as the mixed liquid G used as the learning data. The container 11 containing the liquid mixture G is placed on the turntable 5 after being vibrated under the same uniform conditions. Then, after the vibration is applied, the state of the mixed liquid G in the container 11 after the lapse of a specified time is acquired as image data D. FIG. In the process of photographing the image data D, as illustrated in FIG. 2, the storage unit 3 blocks external light, and the illumination unit 4 is used to generate an estimation model M of the illuminance for the mixed liquid G. The same setting as when the image data D was acquired is made. Also, the photographing background specification is set to the same as when the image data D for generating the estimated model M is acquired. The image data D of the mixed liquid G to be estimated thus acquired is input to the estimation model M and arithmetically processed by the calculation unit 7 to obtain the particle size distribution of the sediment of the mixed liquid G to be estimated. is calculated.

As described above, in each photographing step of acquiring a large number of image data D used as learning data when generating the estimation model M and the image data D used when using the estimation model M, external light is emitted. Each image data D is acquired inside the housing portion 3 in the blocked state. That is, since each image data D is obtained under photographing conditions in which the influence of external light can be substantially ignored, the risk of inclusion of disturbance factors in each image data D is greatly reduced.

Furthermore, in each photographing process, the illumination means 4 is used to set substantially the same illuminance for the object G placed inside the container 3, and the photographing background specifications of the object G are substantially the same. are set to be essentially the same. Therefore, even if each image data D contains a disturbance factor, the variation in the disturbance factor is eliminated or extremely reduced. As a result, it is advantageous to stably acquire the image data D without lowering the estimation accuracy of the estimation model M. In addition to using the respective image data D as they are, for example, when there is a range in which the influence of reflection of the light source cannot be ignored, the data can be processed to remove the range and used as the image data D. Also, in order to improve the estimation accuracy of the estimation model M, the resolution of the image data D can be set higher. However, this requires more memory capacity for storing the image data D and more time for data processing. Therefore, the resolution of the image data D is set to an appropriate level within a range in which the estimation accuracy of the estimation model M can be maintained at a predetermined level.

It is desired that the result of estimation using the estimation model M (particle size distribution of soil) be grasped as early as possible at the construction site. For that purpose, the image data D used when using the estimation model M is acquired at each construction site. A large amount of image data D used when generating the estimation model M is generally acquired at a location different from the image data D used when using the estimation model M. FIG. Therefore, the imaging conditions of the image data D used for generating the estimation model M and the image data D used when using the estimation model M tend to deviate greatly. When the present invention is used, the imaging conditions for each image data D can be made substantially the same, so that highly accurate estimation results by the estimation model M can be grasped in a timely manner at the construction site.

Acquisition system 1 may be a portable unit to facilitate portability. Since the photographing device 2 and the computing unit 7 are easy to carry individually, it is desirable to make at least the storage unit 3 and the illumination means 4 of the acquisition system 1 portable units. If the acquisition system 1 has accessories such as the turntable 5 and the moving mechanism 6, the accessories may also be included in the transportable unit.

When at least part of the acquisition system 1 is made a portable unit in this way, the image data D used for generating the estimation model M and the image data D used when using the estimation model M are acquired at different locations. Even in this case, the acquisition system 1 can be used easily. Along with this, regardless of where the image data D is acquired, the number of disturbance factors included in each image data D is reduced, or the disturbance factors become more uniform. Therefore, it is more advantageous to stably acquire the image data D without lowering the estimation accuracy of the estimation model M.

If the estimation result by the estimation model M matches the actual measurement result corresponding to the estimation by the estimation model M, it can be said that the estimation accuracy of the estimation model M is 100% accurate. However, in reality, it is impossible to achieve 100% estimation accuracy due to various factors. As a result of various investigations by the inventors, several factors that influence the difference between the estimated results and the actual measurement results have been found. Therefore, it is preferable to adjust these factors based on the difference between the two.

For example, the difference between the estimation result and the actual measurement result may change due to the difference in the illuminance inside the housing unit 3 . This is because the difference in illuminance affects the degree of reflection of light on the surface of the object G and the like. Therefore, the illuminance for the object G by the illumination means 4 is changed to an illuminance that reduces the difference between the two. That is, the illuminance is set higher or lower to a more appropriate value. Specifically, the illuminance on the surface of the object G when the image data D is captured is set to a more appropriate value. Under the condition that the illuminance is set to a more appropriate value in this manner, each photographing process is performed to acquire the image data D. FIG.

The difference between the estimation result and the actual measurement result may change depending on the difference in the background color of the object G among the photographing background specifications. This is because the difference in background color affects how light is reflected on the background surface. In the above-described embodiment, the inner surface of the housing portion 3 serves as the background surface when the image data D of the object G is captured, so the background color is changed to reduce the difference between the two. That is, the background color is set to a different color and set to a more appropriate color. In general, the background color is set to have a greater contrast with the color of the object G. Under the condition that the background color is set to a more appropriate color in this way, the image data D is acquired by performing each photographing process.

The difference between the estimated result and the actual measurement result may change depending on the difference in the surface roughness of the background surface among the shooting background specifications. This is because the difference in the surface roughness of the background surface affects how light is reflected on the background surface. Therefore, the surface roughness of the background surface is changed to make the difference between the two smaller. That is, the surface of the background color is made rougher or smoother to set a more appropriate surface roughness. As the surface roughness value, the arithmetic mean roughness Ra specified in JIS B 0601 can be used. Image data D is acquired by performing each photographing process under the condition that the background surface is set to have a more appropriate surface roughness.

In the acquisition system 1 illustrated in FIG. 4, based on the difference between the estimation result and the actual measurement result, the photographing device 2 (image acquisition unit 2a), the lighting means 4 (light emitting unit 4a and diffusion plate 4b), and the rotating table 5 can be adjusted as needed. Therefore, the image data D can be acquired by performing each photographing process under the condition that the relative positions of these components are appropriately set to reduce the difference between the estimation result and the actual measurement result.

In the embodiment of the acquisition system 1 illustrated in FIGS. 6 and 7, the moving mechanism 6 for moving the photographing device 2 is provided, and the object G is placed on the fixed base 10 and placed at a predetermined position inside the container 3. is fixed to The housing portion 3 has a cylindrical shape. The fixed base 10 is preferably installed at the center of the housing portion 3 in plan view. The lighting means 4 are the same as in the previous embodiment.

The moving mechanism 6 has a guide 6b that extends annularly around the fixed table 10 in a plan view, and a moving table 6a that moves along the guide 6b. The photographing device 2 is placed and fixed on a moving table 6a. The moving table 6a is moved at a desired speed by a drive motor or the like. By moving the movable table 6a on which the photographing device 2 is placed, the photographing device 2 rotates around the object G in plan view. The imaging device 2 (image acquisition unit 2a) moves in the circumferential direction of the object G due to the movement of the movable table 6a.

In this embodiment, by photographing the image data D while rotating the photographing device 2 around the object G in plan view, a plurality of images having different positions in the circumferential direction with respect to each object G are obtained. Image data D can be obtained. Since the object G is fixed at a fixed position and does not move, in the case of a liquid object G, it is advantageous to obtain stable image data D without causing the object G to sway due to movement. become.

Further, in this embodiment, since the container 3 has a cylindrical shape, the background of the object G (the inner surface of the container 3) becomes a uniformly curved surface even if the movable table 6a (the photographing device 2) moves. As a result, changes in the specification of the photographing background are avoided, which is more advantageous for obtaining stable image data D. FIG.

In the embodiment of the acquisition system 1 illustrated in FIGS. 8-9, a background plate 9 is installed inside the container 3 . In this embodiment, the surface 9a of the background plate 9, which is separate from the housing portion 3, is used as the background surface. The background board 9 is arranged on the back side of the object G with respect to the photographing device 2, and the object G is arranged between the photographing device 2 and the background plate 9 in plan view. In this embodiment, a background plate 9 having a shape obtained by cutting a cylindrical body in half lengthwise is used to equalize the distance between the outer peripheral surface of the object G and the surface 9a (background surface). It is preferable to use the background plate 9 as wide as possible on the back of the object G. FIG.

Since the background plate 9 is separate from the housing portion 3, it can be easily replaced. Therefore, it is preferable to prepare a plurality of types of background plates 9 having different specifications (color and roughness of the surface 9a). As a result, by selecting and using the background plate 9 with appropriate specifications, it is possible to set shooting background specifications that reduce the difference between the estimation result of the estimation model M and the actual measurement result.

Since the present invention can be applied to obtaining the image data D used when generating and using the estimated model M, the object G is not limited to the liquid mixture described above. From the image data D of the cylindrical boring sample of the ground (for example, the image data D of the side view of the cylinder), the sediment composition and ground strength of the ground can be grasped to some extent. Therefore, it is also possible to apply the present invention when generating the estimation model M for estimating them and when obtaining the image data D used when using it. In this case, the object G is a cylindrical boring sample of the ground.

1 Acquisition System 2 Photographing Device 2a Image Acquisition Unit 3 Storage Unit 3a Main Body 3b Opening/Closing Lid 4 Lighting Means 4a Light Emitting Unit (Light)
4b diffusion plate 4c illuminance meter 4d controller 5 rotating table 6 moving mechanism 6a moving table 6b guide 7 computing unit 8 monitor 9 background plate 9a surface (background surface)
10 Fixed table 11 Container 12 Input layer 13 (13a, 13b, 13c) Intermediate layer 14 Output layer 15 Node G Object (mixture)

In order to achieve the above object, the image data acquisition method of the present invention provides a method for acquiring image data used for generating and using an estimation model by capturing an object of each image data in each capturing step using a capturing device. and at least the image acquisition unit of the photographing device is arranged inside a light-shielding housing section to block light from the outside of the housing section, and the lighting means is used to illuminate the interior of the housing section. The illuminance for the object is set to be the same, the photographing background specifications are set to be the same, and the object is a mixture of liquid and particles contained in a transparent container or a cylindrical boring sample of the ground. It is characterized by
Another method of acquiring image data according to the present invention is a method of acquiring image data used for generating and using an estimation model by photographing an object and the photographing device in each photographing step. at least the image acquisition unit of (1) is placed inside a light-shielding container, and light from the outside of the container is blocked, and the illuminance for the object is obtained by using a lighting means that illuminates the inside of the container. are set to be the same, the shooting background specifications are set to be the same, and based on the difference between the estimation result by the estimation model and the actual measurement result corresponding to the estimation of the estimation model, the background color of the shooting background specification is changed By changing, the background color is set to a color that makes the difference smaller.
Still another image data acquisition method of the present invention is characterized in that, in each image capturing step of capturing and acquiring image data used for generating and using an estimation model with an image capturing device, the object of each of the image data and the image data are captured. At least the image acquisition unit of the apparatus is arranged inside a light-shielding container, and the light outside the container is blocked, and the object is illuminated using illumination means for illuminating the inside of the container. The illuminance is set to be the same, the shooting background specifications are set to be the same, and based on the difference between the estimation result by the estimation model and the actual measurement result corresponding to the estimation of the estimation model, the surface of the background surface among the shooting background specifications The roughness is changed to set the surface roughness to a roughness that makes the difference smaller.

An image data acquisition system according to the present invention includes an imaging device that captures and acquires image data used for generating and using an estimation model, a housing section that has a light shielding property, lighting means that illuminates the inside of the housing section, a computing unit that stores the image data and performs data processing using the image data; In a state in which an object and at least an image acquisition section of the photographing device are arranged inside the housing section to block light from the outside of the housing section, the illuminance for the object is set to be the same by the illumination means. 1. The photographing background specifications are set to be the same, and the object is a mixture of liquid and particles contained in a transparent container or a cylindrical boring sample of the ground.

Claims (7)

In each photographing step of photographing and acquiring image data used for generating and using an estimation model with a photographing device, the object of each of the image data and at least the image acquisition unit of the photographing device are housed in a light-shielding manner. The illuminance for the object is set to be the same using a lighting means that illuminates the interior of the housing portion in a state where the light outside the housing portion is blocked, and the photographic background specifications are the same. A method for acquiring image data, characterized by setting 2. The image data according to claim 1, wherein the illuminance is changed based on a difference between an estimation result by the estimation model and an actual measurement result corresponding to the estimation by the estimation model, and the illuminance is set to make the difference smaller. How to get Based on the difference between the estimation result by the estimation model and the actual measurement result corresponding to the estimation by the estimation model, the background color of the photographing background specifications is changed, and the background color is changed to a color that reduces the difference. 3. The image data acquisition method according to claim 1, wherein the image data is set. Based on the difference between the estimation result by the estimation model and the actual measurement result corresponding to the estimation by the estimation model, the surface roughness of the background surface among the shooting background specifications is changed to reduce the surface roughness. 4. The method for obtaining image data according to claim 1, wherein the roughness is set to a value that is equal to that of the image data. By placing the object on a turntable installed inside the housing unit and photographing the image data while rotating the object on the turntable, 5. The image data acquiring method according to claim 1, wherein a plurality of said image data are acquired at different positions in the circumferential direction. By arranging the entire photographing device inside the housing portion and photographing the image data while rotating the photographing device around the object in a plan view, 6. The image data acquiring method according to claim 1, wherein a plurality of said image data are acquired at different positions in the circumferential direction. A photographing device for photographing and obtaining image data used for generating and using an estimation model, a housing portion having a light shielding property, lighting means for illuminating the inside of the housing portion, and the image data in which the image data is stored. and a computing unit that performs data processing using
In each photographing step in which each of the image data is acquired by the photographing device, an object of each of the image data and at least an image acquisition section of the photographing device are arranged inside the accommodating portion, and the accommodating portion 2. An image data acquisition system, wherein the illuminance for the object is set to be the same by the illuminating means and the photographing background specifications are set to be the same by the illuminating means in a state in which external light is blocked.

Images