Method and device for measuring seedling height

Abstract

The invention discloses a seedling height measuring method and device. The method comprises the steps that S1, a seedling image is collected and preprocessed, and the preprocessed seedling image serves as a to-be-detected target seedling image; s2, performing edge detection on the target seedling image, and extracting a contour map of the target seedling image based on an edge detection result; s3, converting the contour diagram of the target seedling image into a skeleton diagram; and S4, determining the total number of coordinates of the skeleton diagram, and calculating the actual seedling height corresponding to the seedling skeleton diagram by referring to the corresponding relationship between the unit pixel in the scale image and the actual size and the optimal scaling multiple of the scale. The corresponding relation between the unit pixel and the actual size is established, the purpose of measuring the growth height of the wheat seedlings is achieved, and the problems that existing seedling size measurement is high in labor cost, low in accuracy and not universal are solved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of seedling height measurement, and in particular, to a method and device for measuring the height of seedlings. Background Art

[0002] The supervision and measurement of seedling growth involve many aspects of agricultural production. For example, the growth status of seedlings can be obtained in real time, growth abnormalities can be quickly detected, and targeted measures can be taken. Through long-term monitoring data, the growth model of plants can be obtained and a growth model can be established to predict the maturity cycle or assist in the planting plan. In the growth monitoring of different varieties of seedlings, by comparing the growth data of different varieties of seedlings, high-quality varieties with strong stress resistance and high yield can be selected.

[0003] Existing technologies such as the rice seedling detection model and method based on the improved YOLOV3 network disclosed in the patent document CN202010736191.3 are characterized in that: it includes a feature extraction module for extracting multi-scale features of the input rice seedling image to obtain multi-scale seedling feature maps, and a multi-scale prediction module for predicting the position of rice seedlings based on the multi-scale seedling feature maps. This method has problems such as complex input image shooting in the early stage, high labor cost, and low accuracy. In addition, the technical patent document CN202420013174.0 proposes a plant size measurement device, which includes a box body, a moving component, a limiting rod body, a laser ranging sensor, a control unit, and a display unit. Among them, the emitting end of the laser ranging sensor is arranged towards the reflecting part, the reflecting part is used for reflecting laser, the control unit is electrically connected to the laser ranging sensor and the display unit, and the display unit is used for displaying measurement data. However, this device does not have universality, requires the configuration of a limiting rod body and a laser ranging sensor, has a high cost, and a complex operation method. Summary of the Invention

[0004] To solve the problems in the existing technology, the present invention provides a method and device for measuring the height of seedlings to simplify the equipment cost and improve the measurement accuracy.

[0005] In a first aspect, the embodiments of the present invention provide a method for measuring the height of seedlings, including:

[0006] S1. Collect a seedling image and perform preprocessing, and use the preprocessed seedling image as the target seedling image to be detected;

[0007] S2. Perform edge detection on the target seedling image, and extract the contour map of the target seedling image based on the edge detection result;

[0008] S3. Convert the contour map of the target seedling image into a skeleton map;

[0009] S4. Determine the total number of coordinates of the skeleton diagram, refer to the correspondence between the unit pixel and the actual size in the scale image, and the optimal scaling factor of the reference scale, and calculate the actual height of the seedling corresponding to the seedling skeleton diagram.

[0010] Optionally, the seedling image is obtained by the data acquisition module, and the data acquisition module includes a camera, a reference scale, a bracket, and a tray. The camera is vertically arranged above the tray, the reference scale and the seedlings are arranged in the tray, and the camera is used to collect the image information of the seedlings to be detected and the reference scale in the tray.

[0011] Optionally, the S2 specifically includes:

[0012] Use the improved Canny algorithm to perform edge detection on the target seedling image to obtain the edge detection result of the target seedling image;

[0013] Based on the edge detection result, use the findContours() function in OpenCV to extract the structured contour data of the seedlings and establish a contour hierarchy relationship;

[0014] Use the Sorted function to sort the contour data by the outer perimeter, retain the top 10 largest contours for combined drawing and remove the small contours to obtain the contour of the target seedling image.

[0015] Optionally, the S3 specifically includes:

[0016] Use the improved Zhang-Suen thinning algorithm to convert the contour map of the target seedling image into a skeleton diagram by calculating the 4-connected region in the sub-iteration condition.

[0017] Optionally, the S4 specifically includes:

[0018] In the seedling skeleton diagram, count the total number of coordinates forming the skeleton diagram by judging the color value of the pixel coordinates;

[0019] Obtain the optimal scaling factor of the reference scale through template matching and determine the physical size of the unit pixel of the reference scale using the line detection algorithm;

[0020] Calculate the actual plant height of the seedlings according to the product of the total number of coordinates of the skeleton diagram, the optimal scaling factor, and the physical size of the unit pixel.

[0021] In a second aspect, an embodiment of the present invention provides a device for measuring the height of seedlings, and the device includes:

[0022] An acquisition unit, configured to acquire a seedling image and perform preprocessing, and use the preprocessed seedling image as a target seedling image to be detected;

[0023] The contour extraction unit is used to perform edge detection on the target seedling image and extract the contour map of the target seedling image based on the edge detection result;

[0024] The skeleton map extraction unit is used to convert the contour map of the target seedling image into a skeleton map;

[0025] The seedling height calculation unit is used to determine the total number of coordinates of the skeleton map, the correspondence between unit pixels and actual sizes in the reference scale image, and the optimal scaling factor of the reference scale, and calculate the actual seedling height corresponding to the seedling skeleton map;

[0026] Optionally, the seedling image is obtained by the data acquisition module. The data acquisition module includes a camera, a reference scale, a bracket, and a tray. The camera is vertically arranged above the tray. The reference scale and the seedlings are arranged in the tray. The camera is used to collect the image information of the seedlings to be detected and the reference scale in the tray.

[0027] Advantages of the present invention:

[0028] 1. The data acquisition device adopted in the present invention does not require special precision instruments, has few equipment requirements, and the detection process is relatively fast, which can save a large amount of economic costs and labor costs.

[0029] 2. The present invention uses an improved Canny-like algorithm to perform edge detection on the seedling image, dynamically sets high and low thresholds to distinguish strong and weak edges, and enhances the accuracy of edge detection. In addition, in the contour extraction part, based on the optimized edge detection result, a combined algorithm is used to extract the structured contour data of the seedlings through the findContours() function and establish a contour hierarchy relationship. The Sorted function is used to sort the contour data according to the peripheral perimeter, and small noise contours are removed. This combined algorithm effectively improves the accuracy and anti-noise performance of contour extraction;

[0030] 3. The present application optimizes the operation efficiency of the algorithm by improving the pixel connectivity in the Zhang-Suen algorithm, providing a new solution for realizing more efficient skeleton extraction.

[0031] 4. The present application improves the MatchTemplate method by introducing a size variable for scaling the matching template, solves the problem that the traditional method can only match targets of the same size as the template image, and significantly improves the accuracy and confidence of matching. Description of the drawings

[0032] Figure 1 It is a flowchart of a method for measuring the height of seedlings provided by an embodiment of the present invention;

[0033] Figure 2Structural schematic diagram of a data acquisition module provided by an embodiment of the present invention;

[0034] Figure 3 Schematic diagram of rice seedling contour extraction provided by an embodiment of the present invention;

[0035] Figure 4 Comparison chart of edge detection effects of different algorithms;

[0036] Figure 5a Original image of rice seedlings provided by an embodiment of the present invention;

[0037] Figure 5b Result diagram of traditional contour extraction provided by an embodiment of the present invention;

[0038] Figure 5c Result diagram of contour extraction provided by an embodiment of the present invention;

[0039] Figure 6a Schematic diagram of 8-neighborhood in Zhang-Suen thinning algorithm;

[0040] Figure 6b Schematic diagram of 4-neighborhood in the improved Zhang-Suen thinning algorithm;

[0041] Figure 7a Operation time result diagram of applying Zhang-Suen algorithm to wheat seedlings;

[0042] Figure 7b Operation time result diagram of applying the improved Zhang-Suen algorithm to wheat seedlings;

[0043] Figure 8a Skeleton diagram obtained by applying Zhang-Suen algorithm to wheat seedlings;

[0044] Figure 8b Skeleton diagram obtained by applying the improved Zhang-Suen algorithm to wheat seedlings;

[0045] Figure 9 Flow chart of counting the pixel coordinates of rice seedling skeletons provided by an embodiment of the present invention;

[0046] Figure 10 Schematic diagram of a reference scale provided by an embodiment of the present invention;

[0047] Figure 11 Target matching sample diagram provided by an embodiment of the present invention. Specific implementation manner

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. Additionally, it should be noted that for ease of description, only the parts related to the present invention rather than all the structures are shown in the drawings.

[0049] Embodiment

[0050] Figure 1 The following is a flowchart of a method for measuring the height of rice seedlings provided by an embodiment of the present invention, which specifically includes the following steps:

[0051] S1. Collect the rice seedling image and perform preprocessing, and use the preprocessed rice seedling image as the target rice seedling image to be detected.

[0052] Among them, the rice seedling image is collected by a data acquisition module, as Figure 2 shown. The data acquisition module includes a camera 1, a reference scale 2, a bracket 3, and a tray 4. The camera 1 has the same relative distance from the tray 4, and the reference scale 2 has the same relative distance from the tray 4. The camera 1 is used to collect the image information of the rice seedlings to be detected in the tray 4 and the reference scale 2.

[0053] S2. Perform edge detection on the target rice seedling image, and extract the contour map of the target rice seedling image based on the edge detection result.

[0054] Figure 3 The following is a contour extraction flowchart for this embodiment. First, the improved Canny algorithm (Canny-Like) is used to perform edge detection on the target rice seedling image to obtain the edge detection result of the target rice seedling image.

[0055] Among them, the above-mentioned improved Canny algorithm dynamically sets a threshold interval on the basis of the traditional Canny algorithm, with a high threshold and a low threshold. For the high threshold, the pixel points with gray values higher than the high threshold are retained as edge points, and these points usually represent the edges of the image. The pixel points with gray values less than the low threshold are eliminated. For the pixel points within the range of the high and low thresholds, if the pixel point is connected to one or more strong edge points greater than the high threshold, it is retained; otherwise, it is eliminated.

[0056] Figure 4 The following is a comparison chart of the edge detection effects of different algorithms, and the following table shows the edge detection results of different algorithms:

[0057]

[0058] By Figure 4As can be seen from the above table, the method in this embodiment combines the high precision of the Canny algorithm and the rapidity of the Sobel algorithm, while reducing false positives and improving edge continuity, ensuring the response ability to real edges, and also improving edge continuity by connecting strong edge points. Compared with the traditional Canny algorithm, the Canny-like algorithm reduces the computational cost and improves the processing speed while maintaining high precision.

[0059] Optionally, before edge extraction, the image can be denoised and smoothed to further improve the accuracy of edge extraction.

[0060] Then, based on the edge detection results, the findContours() function in OpenCV is used to extract the structured contour data of the seedlings, obtain the coordinate information of all contours, and establish a contour hierarchy relationship;

[0061] Next, the Sorted function is used to sort the contour data by the outer perimeter, retain the top 10 largest contours for combined drawing and remove small contours to obtain the contour of the target seedling image. This combined algorithm effectively improves the accuracy and anti-noise performance of contour extraction. Figure 5a is the original seedling image collected by the camera, and the traditional contour extraction result image is shown in Figure 5b , and the contour extraction result based on the above method can be seen in Figure 5c , it can be seen that this combined algorithm effectively improves the accuracy and anti-noise performance of contour extraction.

[0062] S3. Convert the contour map of the target seedling image into a skeleton map.

[0063] In this embodiment, an improved Zhang-Suen thinning algorithm is used to convert the contour map of the target seedling image into a skeleton map by calculating 4-connected regions in the sub-iteration condition.

[0064] Among them, the Zhang-Suen algorithm is divided into two sub-iteration steps, and the two sub-iteration steps are executed alternately. Each iteration deletes boundary pixels that meet specific conditions until no further thinning is possible. However, through experimental verification, it can be seen that although the Zhang-Suen algorithm can meet the requirements of skeleton extraction, there is still much room for improvement in its operating efficiency. Therefore, this application optimizes the operation efficiency of the algorithm by improving the pixel connectivity in the Zhang-Suen algorithm, providing a new solution for realizing more efficient skeleton extraction.

[0065] First, regarding the common types of connectivity in image processing, there are 4-connectivity and 8-connectivity. The traditional Zhang-Suen thinning algorithm is 8-connectivity. The main idea of the improved Zhang-Suen thinning algorithm in this application is to change the calculation of the 8-connected region in the traditional sub-iteration condition to only consider the 4-neighborhood (P2, P4, P6, P8), and ignore the diagonal neighborhood (P3, P5, P7, P1). See Figure 6a and Figure 6b .

[0066] The constraint condition formula after the improved algorithm is:

[0067] In sub-iteration 1, delete the right and bottom boundary pixels;

[0068] Condition 1: 1 ≤ B4(P1) ≤ 3, the number of foreground pixels in the 4-neighborhood;

[0069] Condition 2: A4(P1) = 1, the number of 0 and 1 transitions in the 4-neighborhood;

[0070] Condition 3: P4 = 0 or P6 = 0, the right or bottom is the background;

[0071] In sub-iteration 2, delete the left and top boundary pixels;

[0072] Condition 1 and Condition 2 are the same as above;

[0073] Condition 3: P2 = 0 or P8 = 0, the left or top is the background;

[0074] Note: B4(P1) = P2 + P4 + P6 + P8, only the 4-neighborhood;

[0075] A4(P1) is the number of 0 and 1 transitions from P2 → P4 → P6 → P8 → P2.

[0076] Figure 7a and Figure 7b are the running times of applying the Zhang-Suen algorithm and the improved Zhang-Suen algorithm to wheat seedlings respectively; Figure 8a and Figure 8b are the skeleton graphs extracted by applying the Zhang-Suen algorithm and the improved Zhang-Suen algorithm to wheat seedlings respectively.

[0077] S4. Determine the total number of coordinates of the skeleton graph, refer to the correspondence between the unit pixel and the actual size in the scale image, and the best scaling factor of the scale, and calculate the actual seedling height corresponding to the seedling skeleton graph.

[0078] First, in the seedling skeleton graph, count the total number of coordinates that make up the skeleton graph by judging the color values of the pixel coordinates.

[0079] The flowchart is as Figure 9 :

[0080] Step 1, traverse the pixels of the skeleton image: Check the color values of each pixel in the skeleton image one by one.

[0081] Step 2, color value judgment: When the color value of the current pixel is greater than 200, it is regarded as valid coordinate data.

[0082] Step 3, count the valid pixels: After the traversal is completed, count the total number of all valid pixels as Points. The total number of pixels Points will be an important parameter for outputting the plant height and is used for subsequent physical size conversion.

[0083] Furthermore, in this embodiment, the template matching Match Template in OpenCV is used to obtain the best scaling factor of the reference scale.

[0084] The core idea of the template matching Match Template in OpenCV is to traverse the target image through a sliding window, calculate the similarity between the target image template and the current template sample window pixel by pixel, so as to find the position of the best matching target. Among them, in this application, the template matching is mainly through the calibration scale sample image such as Figure 10 to match the flexible ruler in the wheat seedling sample image such as Figure 11 of the original color image, so as to achieve the data relationship of converting pixel coordinates into the actual plant height.

[0085] The template matching Match Template requires that the template image and the matching object in the target image have the same size. For example, if the size of the template image is 50×200 pixels, the matching object in the target image also needs to be of the same size. However, the function matchTemplate() may not accurately frame the matching area. To solve this problem, when inputting the scale template, the template size can be adjusted through the resize() function. The adjusted value is determined based on the defined scaling range and the number of loops.

[0086] In this embodiment, the scaling range is set from 0.3 to 1.5 times, and the number of loops is 20 times, which is defined using np.linspace(0.3, 1.5, 20) in the Numpy function library. In the first loop, the target is resized to 0.3 times the original size, and in the second loop, the size is adjusted according to the increment of (1.5 - 0.3) / 20, and so on. By recording the matching results of each loop, the best scaling range is determined. Denote the best scaling ratio of the sample image of this experiment as M, then M = 0.763157, and at this time, the best matching effect can be obtained.

[0087] Furthermore, in this embodiment, the straight line detection algorithm is used to determine the physical size of the reference scale per pixel.

[0088] Exemplarily, in this embodiment, the HoughLinesP line detection algorithm is applied to analyze the reference scale image, and it is found that the physical length of 20 cm on the scale corresponds to 450 pixel points. Accordingly, the physical size I represented by a unit pixel can be calculated as:

[0089]

[0090] In the current reference scale image, the actual physical size represented by each pixel point I is 0.4444 mm.

[0091] The calculation formula for the physical size of the rice seedlings is as follows:

[0092] Height=I×Points×M

[0093] Where M is the calibrated coefficient obtained, which is the best scaling factor returned in template matching, Points is the pixel length or statistical number of the refined skeleton of the wheat seedlings, and Height is the actual plant height of the wheat seedlings.

[0094] Experimental result verification

[0095] In this experiment, the artificial measurement plant height data was compared with the rice seedling height measurement method in this application to improve the plant height data comparison. The specific data is shown in the following table:

[0096]

[0097] The results of multiple groups of data sampling show that the method adopted in this embodiment has higher accuracy compared with artificial measurement. The minimum error is only 0.01 mm, and the average error is 1.0375 mm. This result shows that this solution has reached a relatively high level in measurement accuracy, and at the same time significantly simplifies the measurement process, avoiding counting mistakes and misjudgments caused by visual fatigue in artificial measurement. In addition, this method also provides an image recording function, which is convenient for subsequent reference and verification.

[0098] Furthermore, this embodiment also provides a device for measuring the height of rice seedlings. The device includes:

[0099] An acquisition unit, configured to acquire a rice seedling image and perform preprocessing, and use the preprocessed rice seedling image as the target rice seedling image to be detected;

[0100] A contour extraction unit, configured to perform edge detection on the target rice seedling image and extract the contour map of the target rice seedling image based on the edge detection result;

[0101] A skeleton map extraction unit, configured to convert the contour map of the target rice seedling image into a skeleton map;

[0102] A seedling height calculation unit, which is used to determine the total number of coordinates of the skeleton diagram, the corresponding relationship between the unit pixel and the actual size in the reference scale image, and the optimal scaling factor of the reference scale, and calculate the actual seedling height corresponding to the seedling skeleton diagram;

[0103] Optionally, the seedling image is obtained by a data acquisition module, and the data acquisition module includes a camera, a reference scale, a bracket, and a tray. The camera is vertically arranged above the tray, the reference scale and the seedlings are arranged in the tray, and the camera is used to collect the image information of the seedlings to be detected and the reference scale in the tray.

[0104] Optionally, the contour extraction unit specifically includes:

[0105] Perform edge detection on the target seedling image using an improved Canny algorithm to obtain the edge detection result of the target seedling image;

[0106] Based on the edge detection result, use the findContours() function in OpenCV to extract the structured contour data of the seedlings and establish a contour hierarchy relationship;

[0107] Use the Sorted function to sort the contour data by the outer perimeter, retain the top 10 largest contours for combined drawing and remove small contours to obtain the contour of the target seedling image.

[0108] Optionally, the skeleton diagram extraction unit specifically includes:

[0109] Adopt an improved Zhang-Suen thinning algorithm to convert the contour diagram of the target seedling image into a skeleton diagram by calculating the 4-connected region in the sub-iteration condition.

[0110] Optionally, the seedling height calculation unit specifically includes:

[0111] In the seedling skeleton diagram, count the total number of coordinates forming the skeleton diagram by judging the color value of the pixel coordinates;

[0112] Obtain the optimal scaling factor of the reference scale through template matching and use the line detection algorithm to determine the physical size of the unit pixel of the reference scale;

[0113] Calculate the actual plant height of the seedlings according to the product of the total number of coordinates of the skeleton diagram, the optimal scaling factor, and the physical size of the unit pixel.

[0114] The seedling height measurement device provided by the embodiments of the present invention can execute the seedling height measurement method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0115] Note that the above is only a preferred embodiment of the present invention and the technical principles applied. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments. Without departing from the concept of the present invention, more other equivalent embodiments can be included, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for measuring the height of a seedling, characterized in that, Including: S1. Collect the seedling image and perform preprocessing, and use the preprocessed seedling image as the target seedling image to be detected; S2. Perform edge detection on the target seedling image, and extract the contour map of the target seedling image based on the edge detection result; S3. Convert the contour map of the target seedling image into a skeleton map; S4. Determine the total number of coordinates of the skeleton map, the correspondence between the unit pixel and the actual size in the reference scale image, and the optimal scaling factor of the reference scale, and calculate the actual seedling height corresponding to the seedling skeleton map.

2. The method according to claim 1, characterized in that, The seedling image is obtained by a data acquisition module, and the data acquisition module includes a camera, a reference scale, a bracket, and a tray. The camera is vertically arranged above the tray, the reference scale and the seedlings are arranged in the tray, and the camera is used to collect the seedling image to be detected and the image information of the reference scale in the tray.

3. The method according to claim 1, wherein The specific content of S2 includes: Use an improved Canny algorithm to perform edge detection on the target seedling image to obtain the edge detection result of the target seedling image; Based on the edge detection result, use the findContours() function in OpenCV to extract the structured contour data of the seedlings and establish a contour hierarchy relationship; Use the Sorted function to sort the contour data according to the perimeter of the outer periphery, retain the top 10 largest contours for combined drawing and remove small contours to obtain the contour of the target seedling image.

4. The method according to claim 1, wherein The specific content of S3 includes: Use an improved Zhang-Suen thinning algorithm to convert the contour map of the target seedling image into a skeleton map by calculating the 4-connected region in the sub-iteration condition.

5. The method according to claim 1, wherein The specific content of S4 includes: In the seedling skeleton map, count the total number of coordinates forming the skeleton map by judging the color value of the pixel coordinates; Obtain the optimal scaling factor of the reference scale through template matching and determine the physical size of the unit pixel of the reference scale using a line detection algorithm; Calculate the actual plant height of the seedlings according to the product of the total number of coordinates of the skeleton map, the optimal scaling factor, and the physical size of the unit pixel.

6. A measuring device for the height of seedlings, characterized in that, The device includes: An acquisition unit, which is used to collect the seedling image and perform preprocessing, and use the preprocessed seedling image as the target seedling image to be detected; A contour extraction unit, which is used to perform edge detection on the target seedling image and extract the contour map of the target seedling image based on the edge detection result; A skeleton map extraction unit, which is used to convert the contour map of the target seedling image into a skeleton map; A seedling height calculation unit, which is used to determine the total number of coordinates of the skeleton map, the correspondence between the unit pixel and the actual size in the reference scale image, and the optimal scaling factor of the reference scale, and calculate the actual seedling height corresponding to the seedling skeleton map.

7. The device according to claim 6, characterized in that, The seedling image is obtained by a data acquisition module, and the data acquisition module includes a camera, a reference scale, a bracket, and a tray. The camera is vertically arranged above the tray, the reference scale and the seedlings are arranged in the tray, and the camera is used to collect the seedling image to be detected and the image information of the reference scale in the tray.

Images