Medical imaging device, medical imaging system, medical imaging method, and program

Abstract

To provide an image diagnostic device capable of accurately diagnosing the growth status of plants from images. [Solution] The image diagnostic device includes a first acquisition unit that acquires first and second indicators of a plant using images of the plant, and a second acquisition unit that includes a trained model which has been trained using the first and second indicators, and which acquires at least one of the number of stems and the number of leaves of the plant using the trained model, wherein the first indicator is an indicator relating to at least one of the number of leaves and the size of leaves of the plant, acquired without considering the effect of leaf occlusion, and the second indicator is an indicator relating to leaf occlusion.

Description

Technical Field

[0001] The present invention relates to an image diagnostic apparatus for diagnosing the growth state of a plant using an image of the plant.

Background Art

[0002] In recent years, methods for diagnosing the growth state of a plant using an image of the plant have been proposed. Patent Document 1 discloses a configuration in which the number of leaves is measured using an image of a plant, and the number of stems is obtained based on the correlation between the number of leaves and the number of stems. Further, Patent Document 2 discloses a configuration in which the number of stems can be diagnosed even in a field where the number of leaves cannot be accurately measured by correcting the number of leaves and the number of stems measured from an image of a plant with a vegetation rate.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] An object of the present invention is to provide an image diagnostic apparatus capable of diagnosing the growth state with high accuracy from an image of a plant.

Means for Solving the Problems

[0005] An image diagnostic device as one aspect of the present invention comprises a first acquisition unit that acquires first and second indicators of a plant using an image of the plant, and a second acquisition unit that includes a trained model that has been trained using the first and second indicators, and acquires at least one of the number of stems and the number of leaves of the plant using the trained model, wherein the first indicator is an indicator relating to at least one of the number of leaves and the size of the leaves of the plant, acquired without considering the effect of leaf occlusion, and the second indicator is an indicator relating to the occlusion of the leaves. [Effects of the Invention]

[0006] According to the present invention, it is possible to provide an image diagnostic device that can diagnose the growth state of plants with high accuracy from images. [Brief explanation of the drawing]

[0007] [Figure 1] This flowchart shows a diagnostic imaging method according to an embodiment of the present invention. [Figure 2] This is a conceptual diagram of the learning model used in the leaf count calculation unit. [Figure 3] This diagram shows the training method for the learning model used in the leaf count calculation unit. [Figure 4] This figure shows an example of a training sample for the leaf count calculation unit. [Figure 5] This graph shows the correlation between the number of leaf tips in an image and the measured number of stems, in the absence of occlusion effects. [Figure 6] This diagram illustrates the complex occlusion of leaves caused by adjacent plants. [Figure 7] This graph shows the effect of complex leaf occlusion by adjacent plants on stem count calculation. [Figure 8] This diagram explains the cause of occlusion at the leaf tip. [Figure 9] This figure shows examples of captured images and leaf region images. [Figure 10] This is a conceptual diagram of the learning model used in the leaf region calculation unit. [Figure 11]This diagram illustrates an example of the calculation range for vegetation cover calculated by the vegetation cover calculation unit. [Figure 12] This is a conceptual diagram of the height histogram calculated by the height histogram calculation unit. [Figure 13] This diagram shows the relationship between rice plant height and plant height. [Figure 14] This diagram illustrates how the shape of the height histogram changes depending on the shape of the rice leaf. [Figure 15] This figure shows overlaid height histograms for different leaf shapes. [Figure 16] This figure shows an example of a neural network used in the stem number estimation section. [Figure 17] This is a hardware configuration diagram of a medical imaging device. [Modes for carrying out the invention]

[0008] Embodiments of the present invention will be described in detail below with reference to the drawings. In each drawing, the same reference numeral is used for identical components, and redundant descriptions are omitted.

[0009] In the configuration of Patent Document 1, leaf tips are detected from the image, but the leaf tips of one plant may be obscured by the leaves of an adjacent plant, making accurate detection impossible. In addition, in the configuration of Patent Document 2, the correction start point is determined from a sharp drop in the leaf count or stem count graph, which can lead to a delay in the start of correction, or it may not be able to handle cases where the leaf count or stem count declines gradually without a sharp drop.

[0010] In this embodiment, as an example of an image diagnostic method for diagnosing the growth status of plants where leaf occlusion occurs in densely planted fields where plant leaves are hidden, a method for diagnosing the number of stems from images of rice plants in a field is described. In rice cultivation, the number of stems correlates with the number of panicles at the ripening stage and is therefore used as an indicator for managing rice yield. Note that the subject of this invention is not limited to rice. Other plants include, for example, wheat.

[0011] Figure 1(a) is a flowchart showing an image diagnosis method according to an embodiment of the present invention.

[0012] In step S10, the image input unit reads a photographed image and a distance image of rice.

[0013] In step S20, the leaf count calculation unit calculates the number of leaves from the input rice image. Specifically, the leaf count calculation unit calculates the number of leaves using a trained model generated by machine learning. First, the leaf count calculation unit outputs a density estimation image from the photographed image. As the density estimation image, for example, a Gaussian distribution having a maximum at a position where an object is likely to exist is used. When objects overlap in the image, a density distribution in which density distributions corresponding to each object are superimposed is obtained. As the trained model used by the leaf count calculation unit, a trained model 202 generated by a CNN that solves a regression problem and uses the photographed image as input data 201 and the density estimation image as output data 203 as shown in FIG. 2 is used.

[0014] The photographed image illustrated in FIG. 2 is in monochrome display, but an RGB image is used in actual image diagnosis. When detecting leaves, a monochrome image in the absorption wavelength band of chlorophyll such as a wavelength of 700 nm may be used, but since detection accuracy is improved by using color information, an image including wavelength information of two or more bands may be used.

[0015] When learning by CNN, as shown in FIG. 3, a learning sample 310 is used in which the input data 311 is a rice photographed image and the teacher data 312 is a labeling image labeled with a Gaussian distribution at the leaf tip position. Thereby, it can be output as information representing the density distribution of the object. At this time, when the pixel value of the labeling image is represented by a finite value such as 8 bits, normalization is performed according to the assumed number of overlapping objects. For normalization, for example, when the maximum number of overlapping objects to be assumed is 3, the maximum value of the Gaussian distribution labeled for the object is set to 85, so that the information amount of the labeling image can be utilized to the maximum extent.

[0016] In neural network training, first, an error is calculated using a loss function based on the difference between the output data output from the output layer (which is based on the input data input to the input layer) and the training data. Next, the connection weight coefficients between nodes of the neural network are updated to minimize the obtained error. This update is performed, for example, by using backpropagation to adjust the connection weight coefficients between nodes of the neural network in order to minimize the above error.

[0017] Figure 4 shows an example of a training sample for the leaf count calculation unit, illustrating an example of input data 401 and training data 402 when the maximum number of overlapping labels is set to 3. As shown in Figure 4, Gaussian distribution labels are applied to the positions in training data 402 corresponding to the leaf tips of input data 401. In cases where three leaf tips are clustered together, such as in the dense leaf tip area 403, three Gaussian distributions are superimposed.

[0018] Next, the leaf count calculation unit calculates the number of leaf tips from the output density estimation image. If the density distribution corresponding to one object is normalized so that the sum of densities is 1, the sum of the density estimation images corresponds to the number of objects in the image. For example, if an image like the training data 402 in Figure 4 is input as the density estimation image, there are approximately 12 Gaussian distributions, so the number of leaf tips will be approximately 12.

[0019] Note that the range used to calculate the number of leaf tips does not have to be the entire image, but rather a portion of the image.

[0020] Furthermore, while density estimation images are used to calculate the number of leaves in this embodiment, the present invention is not limited to this. Leaf tips may be detected and the number of leaves calculated using methods such as pattern recognition techniques like pattern matching, leaf tip detection using local features such as SIFT features, or leaf tip detection by segmentation using machine learning.

[0021] In this embodiment, the number of leaves is calculated, but the number of stems may also be calculated. Figure 5 is a graph showing the correlation between the number of leaf tips in an image and the measured number of stems when there is no influence from occlusion. The horizontal axis represents the number of leaf tips for 10 plants in the image, and the vertical axis represents the average measured number of stems for 10 plants. In Figure 5, the coefficient of determination is very high at 0.95, indicating a strong correlation between the number of leaves and the number of stems, so this correlation may be used to convert the number of leaves to the number of stems. In this case, as shown in Figure 1(b), the processing in step S21 is performed by the stem number conversion unit after step S20. However, the number of stems calculated in step S21 differs from the value estimated by the stem number estimation unit described later, as it is affected by occlusion. In addition, both the number of leaves and the number of stems may be calculated.

[0022] Figure 6 illustrates the complex occlusion of leaves by adjacent plants, and shows a composite image 601 of a captured image and a density-estimated image. In Figure 6, there are three plants in the image, and the leaves of adjacent plants overlap in a complex manner. By using the density-estimated image, it is possible to detect leaf tips if the occlusion 603 is simple enough for a human to recognize that the leaf tip is hidden by another leaf. However, it is impossible even manually to recognize all leaf tips present in region 602. Occlusion by leaves of adjacent plants becomes more complex as the rice grows.

[0023] Figure 7 is a graph showing the effect of complex leaf occlusion by adjacent plants on stem count calculation. The horizontal axis represents the date, and the vertical axis represents the stem count converted from the leaf count (circle plot) and the measured stem count (triangle plot). In field 702, which has a low planting density, the stem count converted from the leaf count and the measured stem count are close in value. On the other hand, in field 701, which has a high planting density, the occlusion by leaves of adjacent plants becomes more complex as the rice grows, so the stem count converted from the leaf count is smaller than the measured stem count.

[0024] In this embodiment, the reduction in the calculated stem count due to occlusion is solved by using a pre-trained model generated by machine learning, which is trained using information on leaf count, vegetation cover, plant height, and leaf shape as input.

[0025] Figure 8 illustrates the cause of leaf tip occlusion. From the camera 801 located above, the leaf tips 802 in the upper layer can be seen in the captured image, but the leaf tips 803 in the lower layer may be difficult to see in the captured image due to occlusion. However, the amount of occlusion is complex and depends on factors such as the density of the leaves, the size of the plant being photographed, and the shape (flexibility) of the leaves, making it difficult to express with a simple formula.

[0026] Therefore, in this embodiment, by using a pre-trained model that takes vegetation cover, plant height, leaf shape, and leaf number as inputs and outputs stem number as parameters representing the amount of occlusion, it is possible to accurately estimate the number of stems even when there is leaf occlusion. Note that the vegetation cover is affected by occlusion as the foliage is denser and the lower leaves are affected. The plant height is affected by occlusion as the plant is taller and the lower leaves are affected. The leaf shape is affected by occlusion as the more bent the leaf, the greater the effect of occlusion because the lower leaves are hidden by the upper leaves.

[0027] In step S30, the leaf region calculation unit calculates a leaf region image 902 from the input rice paddy image 901, as shown in Figure 9. To calculate the leaf region image, it is possible to use a pre-trained model generated by machine learning, similar to the leaf count calculation unit. For example, as shown in Figure 10, this can be achieved by using a CNN-based learning model 1002 that has been trained using rice paddy image data 1001 as input data and leaf region images 1003 as training data. For the training data 1003, for example, images in which leaf regions are labeled with 1 and non-leaf regions are labeled with 0 may be used. However, other methods for extracting leaf regions may be used, such as thresholding using vegetation indices that focus on leaf color, such as NDVI or RGBVI, rather than machine learning.

[0028] In step S40, the vegetation cover calculation unit calculates the vegetation cover rate as an indicator representing the density of rice plants from the leaf area calculated by the leaf area calculation unit. The leaf area image created by the leaf area calculation unit is used to calculate the vegetation cover rate. The vegetation cover rate can be calculated by specifying a range within the leaf area image and calculating the leaf area ratio. However, other methods may be used to calculate the vegetation cover rate within a specified range. Alternatively, the ratio of the leaf area in the entire image may be calculated without specifying a calculation range. Furthermore, multiple vegetation cover rates may be calculated. As shown in Figure 11, the calculation range of the leaf area may be specified in the row direction 1101 and the inter-plant direction 1102 perpendicular to it, respectively.

[0029] In step S50, the height histogram calculation unit generates a leaf region height histogram from the leaf region and distance image calculated by the leaf region calculation unit. Specifically, the height histogram calculation unit generates a leaf region height histogram by masking the distance image using the leaf region image extracted by the leaf region calculation unit. The resolution of the height histogram created here should be low for processing by the grass height calculation unit and leaf shape calculation unit described later. Figure 12 is a conceptual diagram of the height histogram calculated by the height histogram calculation unit. In Figure 12, the left side is the sky side and the right side is the ground side. In Figure 12, the low-resolution height histogram 1201 and the high-resolution height histogram 1202 are histograms generated from the same captured image, distance image, and leaf region image. When the grass height calculation unit and leaf shape calculation unit calculate the peak value of the height histogram, the high-resolution height histogram 1202 may have multiple peaks, potentially causing errors. Therefore, it is preferable to generate a height histogram with a limited resolution, as shown in low-resolution height histogram 1201. Alternatively, a height histogram can be generated with a finer resolution, and the generated height histogram can be smoothed. Furthermore, in this height histogram, large or small values ​​in the distance image that cannot be considered leaf regions may be removed by thresholding.

[0030] In step S60, the plant height calculation unit calculates the plant height from the leaf region height histogram generated by the height histogram calculation unit and converts it to plant height. The peak value of the height histogram is used as the plant height. Figure 13 shows the relationship between rice plant height and plant height. Plant height 1301, an indicator of the growth state of rice, is the height of the plant with the leaves straightened, but since plants such as rice have curvature, it is close to the plant height 1302 that can be measured in the distance image. Therefore, the conversion to plant height uses the height representative value and the conversion coefficient for plant height. When the plant height (plant height) is H and the conversion coefficient is α, the plant height L can be expressed by the following equation (1).

[0031] L = H × α (1) In this embodiment, the peak value of the height histogram is used for the plant height, but other values ​​such as the mean or median of the height histogram may also be used. The conversion coefficient α from plant height to plant height varies depending on the shape of the leaves, so it may be changed according to values ​​representing the amount of leaf deflection, such as the histogram distance or histogram variance, calculated by the leaf shape calculation unit described later.

[0032] In this embodiment, the final output is calculated as plant height, but it is also possible to output another value representing the height of the rice plants, such as plant height, or to calculate both plant height and plant height.

[0033] In step S70, the stem shape calculation unit calculates values ​​related to leaf shape from the height histogram of the leaf region calculated by the height histogram calculation unit. Figure 14 illustrates the change in the shape of the height histogram due to the shape of the rice leaves, and shows examples of generating histograms related to the height of plants for plant 1401 with a small amount of leaf deflection and plant 1405 with a large amount of deflection. In the case of plant 1401, the histogram 1402 has a shape that spreads out to the left and right because the amount of deflection is small. At this time, the mode 1403 of histogram 1402 corresponds to the middle of the plant, and the distance 1404 between the mode 1403 and the tip of the leaf is large. In the case of plant 1405, the histogram 1406 has many high values ​​close to the camera because the amount of deflection is large. Therefore, the mode 1407 of histogram 1406 corresponds to a position higher than the middle of the plant, and the distance 1408 between the mode 1407 and the tip of the leaf is small. Figure 15 shows histograms 1402 and 1406 superimposed. As shown in Figure 15, the magnitude of leaf deflection is expressed in the distance between the mode and the tip of the leaf (hereinafter referred to as the histogram distance).

[0034] The leaf shape calculation unit calculates the histogram distance, which is the difference between the leftmost value and the mode of the height histogram. When detecting the leftmost value of the histogram, it is best to detect the highest point (the point closest to the camera) where there are a certain number of data points, such as 1000 or more, in order to reduce the influence of erroneous distance measurements due to noise and other factors. The shorter the histogram distance, the greater the amount of deflection, so the upper leaves are more likely to hide the lower leaves, and the effect of occlusion at the leaf tips is greater.

[0035] In this embodiment, the histogram distance is calculated, but other values ​​that change depending on the amount of leaf deflection, such as the width, kurtosis, and variance of the histogram, may also be calculated, or both the histogram distance and the histogram variance may be calculated.

[0036] In step S80, the stem number estimation unit estimates the number of stems using a trained model that takes as input the values ​​representing the number of leaves, the percentage of vegetation, the height, and the leaf shape calculated by the leaf number calculation unit, the vegetation cover calculation unit, the plant height calculation unit, and the stem shape calculation unit.

[0037] Figure 16 shows an example of a neural network used in the stem count estimation unit. The pre-trained model used in this embodiment can be realized by using a pre-trained neural network (NN) that has been trained to solve a regression problem with the input 1601 being the number of leaves, vegetation cover, plant height, histogram distance, and histogram variance, and the output 1602 being the number of stems. In Figure 16, the network configuration has two hidden layers 1603 and a bias 1604 in each layer, but the network configuration is not limited to this, and other configurations may be used.

[0038] In this embodiment, the inputs were leaf count, vegetation cover, plant height, and histogram information (histogram distance and histogram variance). However, as mentioned above, the inputs are not limited to these; plant height and stem count may also be included. Furthermore, planting density and plant spacing may be added to the inputs, or at least two inputs may be used, such as inputting vegetation cover as an indicator representing the number and size of leaves (first indicator) and plant height as an indicator representing the amount of leaf occlusion (second indicator). In addition, different vegetation cover rates may be input, such as inputting the vegetation cover in the row direction 1101 in Figure 11 as an indicator representing the number and size of leaves, and the vegetation cover in the inter-plant direction 1102 as an indicator representing the amount of leaf occlusion. As mentioned above, the amount of leaf occlusion is intricately related to factors such as the density of growth, the size of the plant being photographed, and the shape of the leaves (flexibility). Therefore, using multiple indicators representing the amount of leaf occlusion as inputs can be expected to produce a more accurate output of the stem count.

[0039] Note that the calculation of indicators not used for input in step S80 may be omitted from the flow in Figure 1.

[0040] Furthermore, while the output is currently set to the number of stems, the output could also be set to the number of leaves, taking into account the amount of occlusion.

[0041] As shown in Figure 5, there is a strong correlation between the number of leaves and the number of stems in an image under conditions where occlusion does not affect the image. Therefore, instead of using the number of stems, the user can be presented with a value equivalent to the number of stems by converting it to the number of leaves under conditions where occlusion does not affect the image. In this case, the output number of leaves may be presented to the user as is, or it may be converted to the number of stems before being presented.

[0042] Furthermore, the trained models used in this step may be switched between two or more models depending on the conditions. For example, models may be switched depending on the time period, such as using models trained separately for one month after planting and for the period thereafter. Alternatively, if distance images are not captured, a model that only takes leaf count and vegetation cover as input may be used, and the models may be switched based on the calculated indicators. Different trained models may also be used depending on the rice variety and planting density of the target field.

[0043] Furthermore, each indicator, such as the number of leaves and vegetation cover, may be calculated from separate images. For example, to reduce false detections due to shadows, the number of leaves could be calculated using images taken in the early morning, while the vegetation cover could be calculated using images taken at night with a separate light source to minimize the effects of color changes due to ambient light. Images taken under different conditions may be used for each indicator.

[0044] Figure 17 is a hardware configuration diagram of an image diagnostic device 1700 according to an embodiment of the present invention. The image diagnostic device 1700 includes a system bus 1701, a CPU 1702, a ROM 1703, a RAM 1704, an HDD 1705, a NIC 1706, an input unit 1707, a display unit 1708, and a GPU 1709. The image diagnostic system is composed of the image diagnostic device 1700 and a camera (image acquisition device) not shown. The camera acquires images of rice plants. The image diagnostic device 1700 receives the images acquired by the camera via the internet and acquires either the number of stems or the number of leaves using the flow shown in Figure 1.

[0045] The input unit 1707 receives images of rice plants taken with a camera or the like. The ROM 1703 is a storage device that stores the images. The images may be stored in the storage device HDD 1805, or in an external storage device 1711 via the network 1710. The RAM 1704 reads the images stored in the ROM 1703 or HDD 1705. If the images are stored in the external storage device 1711, the RAM 1704 reads the images via the NIC 1706.

[0046] CPU1702 executes the flow shown in Figure 1. CPU1702 functions as a first acquisition unit that obtains indices representing the number and size of plant leaves and an index representing the amount of occlusion of plant leaves, and also functions as a second acquisition unit that obtains either the number of stems or the number of leaves using these indices. Since the estimation of density estimation images in the leaf count calculation unit and the processing in the leaf region calculation unit and stem count estimation unit require CNN or NN calculations, it is preferable to configure the system to execute these on GPU1709. When performing these processes, GPU1709 is used in addition to CPU1702. GPU1709 can perform calculations efficiently by processing more data in parallel, so it is effective to process CNN or NN calculations on GPU1709.

[0047] The display unit 1708 displays the estimated number of leaves and stems. The display unit 1708 may also display information such as the number of leaves and stems up to the present time, a graph of the number of leaves and stems, captured images, density estimation images, leaf region images, height histograms, various indicators, variety, accumulated temperature, and data from the previous year. Furthermore, the displayed number of leaves and stems may show both cases, including and excluding the effects of occlusion, or only one of them.

[0048] As described above, according to the configuration of this embodiment, even in densely planted fields where leaves are hidden, it is possible to diagnose the growth status of plants with high speed and accuracy from images. [Other embodiments] The present invention can also be realized by supplying a program that implements one or more of the functions of the above-described embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. It can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0049] This embodiment includes the following configurations and methods. (Composition 1) A first acquisition unit that acquires first and second indicators of the plant using images of the plant, The system includes a trained model that has been trained using the first and second indicators described above, and a second acquisition unit that uses the trained model to acquire at least one of the number of stems and the number of leaves of the plant, The first indicator is an indicator relating to at least one of the number of leaves and leaf size of the plant, obtained without considering the effect of leaf occlusion. The imaging diagnostic device is characterized in that the second indicator is an indicator relating to the occlusion of the leaf. (Configuration 2) The image diagnostic apparatus according to configuration 1, characterized in that the first indicator includes at least one of the number of leaves and stems, and the vegetation cover rate, which are obtained without considering the effect of leaf occlusion. (Composition 3) The image diagnostic apparatus according to configuration 1 or 2, characterized in that the second indicator includes at least one of the vegetation cover rate, information regarding the height of the plants, information regarding the shape of the leaves of the plants, and information regarding the planting density of the plants. (Composition 4) The image diagnostic device according to configuration 3, characterized in that the information relating to the height of the plant includes at least one of the plant's height and plant height. (Composition 5) The image diagnostic device according to configuration 3 or 4, characterized in that information regarding the shape of the plant's leaves is estimated from a histogram of information regarding the plant's height. (Composition 6) The image diagnostic apparatus according to configuration 5, characterized in that information regarding the shape of the plant leaves is estimated using at least one of the distance from the edge to the peak of the histogram, the width of the histogram, the kurtosis of the histogram, and the variance of the histogram. (Composition 7) The image diagnostic apparatus according to any one of configurations 1 to 6, characterized in that the pre-trained model is obtained by a neural network. (Composition 8) The image diagnostic device according to configuration 7, characterized in that the trained model is obtained by training with at least two of the following as input data: the number of leaves of the plant, the plant's vegetation cover, the plant's height, and the shape of the plant's leaves. (Composition 9) The image diagnostic device according to configuration 7, characterized in that the trained model is obtained by training using at least three pieces of input data: the number of leaves of the plant, the plant's vegetation cover, the plant's height, and the shape of the plant's leaves. (Composition 10) The image diagnostic device according to configuration 7, characterized in that the trained model is obtained by training using information regarding the number of leaves of the plant, the plant's vegetation cover, the plant's height, and the shape of the plant's leaves as input data. (Composition 11) The image diagnostic device according to configuration 7, characterized in that the trained model is obtained by training using as input data information histogram information regarding the number of leaves of the plant, the plant cover rate, the plant height, and the height of the plant. (Composition 12) The image diagnostic apparatus according to any one of configurations 1 to 11, characterized in that the first acquisition unit acquires the first and second indicators using images that are different from each other. (Composition 13) The image diagnostic device according to any one of configurations 1 to 12, characterized in that the aforementioned image is obtained by photographing the plant from above. (Composition 14) The imaging diagnostic device according to any one of configurations 1 to 13, characterized in that the plant includes at least one of rice and wheat. (Composition 15) An image acquisition device that acquires images of plants, A medical imaging system characterized by having a medical imaging device described in any one of configurations 1 to 14. (Method 16) A step of obtaining first and second indicators of the plant using images of the plant, The method includes a trained model that has been trained using the first and second indicators described above, and the step of obtaining at least one of the number of stems and the number of leaves of the plant using the trained model, The first indicator is an indicator relating to at least one of the number of leaves and leaf size of the plant, obtained without considering the effect of leaf occlusion. The image diagnostic method is characterized in that the second indicator is an indicator relating to the occlusion of the leaf. (Composition 16) A program characterized by performing the image diagnostic method described in Method 1.

[0050] Although preferred embodiments of the present invention have been described above, the present invention is not limited to these embodiments, and various modifications and changes are possible within the scope of its gist. [Explanation of Symbols]

[0051] 1700 Medical imaging equipment 1702 CPU (First acquisition unit, Second acquisition unit)

Claims

1. A first acquisition unit that acquires first and second indicators of the plant using images of the plant, The system includes a trained model that has been trained using the first and second indicators described above, and a second acquisition unit that uses the trained model to acquire at least one of the number of stems and the number of leaves of the plant, The first indicator is an indicator relating to at least one of the number of leaves and the size of leaves of the plant, obtained without considering the effect of leaf occlusion. The imaging diagnostic device is characterized in that the second indicator is an indicator relating to the occlusion of the leaf.

2. The imaging diagnostic apparatus according to claim 1, characterized in that the first indicator includes information on at least one of the number of leaves and stems, and the vegetation cover rate, obtained without considering the effects of leaf occlusion.

3. The image diagnostic apparatus according to claim 1 or 2, characterized in that the second indicator includes at least one of the vegetation cover rate, information relating to the height of the plants, information relating to the shape of the leaves of the plants, and information relating to the planting density of the plants.

4. The image diagnostic device according to claim 3, characterized in that the information relating to the height of the plant includes at least one of the plant's height and grass height.

5. The image diagnostic device according to claim 3, characterized in that information regarding the shape of the leaves of the plant is estimated from a histogram of information regarding the height of the plant.

6. The image diagnostic apparatus according to claim 5, characterized in that information regarding the shape of the leaves of the plant is estimated using at least one of the distance from the edge to the peak of the histogram, the width of the histogram, the kurtosis of the histogram, and the variance of the histogram.

7. The image diagnostic apparatus according to claim 1 or 2, characterized in that the trained model is obtained by a neural network.

8. The image diagnostic device according to claim 7, characterized in that the trained model is obtained by training with at least two of the following as input data: the number of leaves of the plant, the vegetation cover rate of the plant, the height of the plant, and the shape of the leaves of the plant.

9. The image diagnostic device according to claim 7, characterized in that the trained model is obtained by training with at least three pieces of input data: the number of leaves of the plant, the plant's vegetation cover, the plant's height, and the shape of the plant's leaves.

10. The image diagnostic device according to claim 7, characterized in that the trained model is obtained by training using information regarding the number of leaves of the plant, the plant's vegetation cover, the plant's height, and the shape of the plant's leaves as input data.

11. The image diagnostic device according to claim 7, characterized in that the trained model is obtained by training using as input data information histogram information regarding the number of leaves of the plant, the plant cover rate, the plant height, and the height of the plant.

12. The image diagnostic apparatus according to claim 1 or 2, characterized in that the first acquisition unit acquires the first and second indicators using images that are different from each other.

13. The image diagnostic device according to claim 1 or 2, characterized in that the image is obtained by photographing the plant from above.

14. The imaging diagnostic apparatus according to claim 1 or 2, characterized in that the plant includes at least one of rice and wheat.

15. An image acquisition device that acquires images of plants, A medical imaging system characterized by having a medical imaging apparatus according to claim 1 or 2.

16. A step of obtaining first and second indicators of the plant using images of the plant, The method includes a trained model that has been trained using the first and second indicators described above, and the step of obtaining at least one of the number of stems and the number of leaves of the plant using the trained model, The first indicator is an indicator relating to at least one of the number of leaves and the size of leaves of the plant, obtained without considering the effect of leaf occlusion. The image diagnostic method is characterized in that the second indicator is an indicator relating to the occlusion of the leaf.

17. A program characterized by performing the image diagnostic method described in claim 16.

Images