Research on the Detection Method of the Threshing Rate of Corn Ears Based on Machine Vision

Abstract

The threshing rate is one of the important indexes to evaluate the effect of corn threshing. The weighing method is often used to calculate the depuration rate of maize at present. This method is time-consuming and laborious and can only calculate the overall threshing rate but does not give the threshing rate of individual corn ears. Different parameters of corn ears have complex effects on the threshing rate. By analyzing the threshing rate of each corn ear, we can choose the appropriate ear treatment method, optimize the processing equipment and process flow, and improve the threshing performance. This paper presents a method based on machine vision to detect the threshing rate of corn ears. In this method, machine vision was used to measure the parameters of the corncob and the area of the top of residual kernels. The area of the top of all kernels was restored based on the parameters of the corncob. The threshing rate of corn ears was calculated by the ratio of the area of the top of the missing kernel to the area of the top of all kernels after threshing. A bivariate linear regression area model was established to restore the area of the top of all corn kernels based on corncob parameters. The R² was more significant than 0.98, and the goodness of fit was good. The machine vision inspection results showed that the maximum relative error of length and midsection radius was 7.46% and 5.55%, and the mean relative error was 2.58% and 2.23%. The maximum relative error of the corn ear threshing rate was 7.08%, and the mean relative error was 2.04%. When the residual kernels were concentrated in the midsection, the inspection result of the corn ear threshing rate was better. The maximum relative error was 3.98%, and the mean relative error was 1.07%. This paper provides a new idea and reference for measuring the threshing rate of corn ears.

1. Introduction

Corn is one of the world’s three major food crops and a vital source of high-quality feed, pharmaceuticals, and raw chemical materials. According to China’s National Bureau of Statistics, the country’s corn planting area reached 430.71 million hectares in 2022, with a total output exceeding 222 million tons. Corn threshing is an essential part of the mechanization of corn production, and it is also the most complex and tedious operation process [1,2,3,4]. The depuration rate is one of the critical indicators to evaluate the effect of corn threshing, which refers to the ability of the threshing machinery to separate the corn kernels from the ear during the corn threshing process, usually expressed as a percentage. If the threshing rate is too low, the corn must be threshed again, which will increase the production cost and the kernel’s breakage rate to a certain extent, resulting in certain economic losses [5,6]. When calculating the corn threshing rate, there is a weighing method: the threshing rate is calculated by the ratio of the weight of the removed kernels to the weight of all kernels [7]. There is also a counting method: the threshing rate is calculated by the ratio of the number of kernels removed to the total number of kernels [8]. These methods are time-consuming and laborious and can only calculate the overall threshing rate and cannot give the threshing rate of individual corn ears. Different parameters of corn ears have complex effects on the threshing rate. By analyzing the threshing rate of each corn ear, we can choose the appropriate ear treatment method, optimize the processing equipment and process flow, and improve the threshing performance. After threshing, the corn ears can be categorized based on the extent of the threshing rate of each corn ear in order to select the appropriate reprocessing method, thereby enhancing raw material utilization and reducing costs. Currently, there is no convenient and quick method for determining the threshing rate of a single ear of corn. However, machine vision has been widely used to detect corn pests, diseases, and loss rates during harvesting and threshing [9,10]. Furthermore, machine vision technology is often used to detect phenotypic parameters of crops quickly [11,12,13], which can help detect the threshing rate of corn ears.

The rapid development of machine vision technology includes 3D point cloud technology and 2D plane imaging technology [14,15,16,17]. Three-dimensional point cloud technology can obtain high-precision spatial data and has the advantages of no image distortion and not being affected by perspective distortion [18,19]. Gené-Mola et al. [20] used SfM-MVS point clouds to measure apple size and estimate apple visibility automatically. The minimum mean absolute error of the measurements was 3.7 mm. Su et al. [21] used a depth camera to obtain 3D shape information of potatoes to measure parameters such as length, width, thickness, and volume to realize automatic potato grading. However, 3D point cloud technology requires a large amount of data storage space, and its complex data processing process makes the processing speed slow. Compared with 3D point cloud technology, although 2D imaging technology has image distortion and other problems, its equipment cost is low, its processing speed is fast, and it is easier to implement. Furthermore, 2D imaging techniques have made distortions that can be corrected by calibration and other methods [22]. Two-dimensional imaging techniques have been widely used to measure phenotypic parameters of crops [23]. Liao et al. [24] used the ellipse fitting algorithm to process 2D images of potatoes. They measured the major axis and minor axis of the potato by setting the correction factor, and the relative error was less than 4%. Tian et al. [25] calculated seedling diameter by processing 2D images of grafted vegetable seedlings. Shen et al. [26] developed an MV-based algorithm to process 2D images of potatoes and estimate their shape and size. Butters et al. [27] and Wang et al. [28] used ellipse fitting to process 2D images and estimate the diameters of apples and mangoes to estimate their sizes, respectively. Huynh et al. [29] and Tran et al. [30] estimated the mass and volume of fruits and vegetables by processing the top-down 2D floor plans of fruits and vegetables and combining linear regression. These studies use 2D imaging techniques to measure the phenotypic parameters of crops quickly and accurately. On this basis, the phenotypic parameters of corn can be measured by 2D imaging equipment to calculate the threshing rate of corn ears.

In order to realize the rapid and automatic measurement of the corn ear threshing rate, this paper proposes a method for detecting the corn ear threshing rate based on machine vision. In this method, the threshing rate of corn ears was calculated by the ratio of the area of the top of the missing kernel to the area of the top of all kernels after threshing. Firstly, the mathematical relationship between the diameter, radius, and length of the corncob and the area of the top of all kernels was analyzed, and the regression model for restoring the area of the top of all corn kernels was established. Then, the parameters of the corncob and the area of the top of the residual kernels were measured by machine vision to detect the threshing rate of a single corn ear. Finally, the accuracy of the detection results was verified by comparing them with the threshing rate of corn ears measured by the weighing method.

2. Materials and Methods

2.1. Experimental Methods

The corn variety used in this paper is Boyun 88. The corn samples were in the full ripening stage, with water content ranging from 29.85 to 32.62%, and were collected manually from the experimental field of Henan University of Science and Technology on 31 October 2023. The ratio of the area of the top of the missing kernels after threshing to the area of the top of all kernels was the threshing rate of the corn ear, as shown in Equation (1):

$$\rho =\frac{S-S_0}{S}\times 100\mathrm{\%}$$

(1)

where $\rho$ is the threshing rate of the corn ear, $S$ is the area of the top of all kernels, and $S_0$ is the area of the top of residual kernels on the corncob after threshing.

The steps are as follows: (1) The area of the top of all corn kernels, the length of corncobs, and the diameter and radius of each segment were counted. After analysis, the appropriate corncob parameters were selected to establish a regression model to restore the area of all corn kernels. (2) The 2D image of the corncob is processed, and the required parameters of the corncob and the area of the top of the residual kernels on the corncob are extracted to calculate the threshing rate of the corn ear. (3) Compare and analyze the results of the corncobs’ automatically measured phenotypic parameters with the actual values to test the detection accuracy. (4) Compare and analyze the threshing rate of the corn ear measured automatically with the value measured by the weighing method to verify the accuracy of the results.

The steps of the weighing method to measure the removal rate of corn ears are as follows: (1) Measure the weight of the corn ear before threshing. (2) Measure the weight of the corncob with residual kernels after threshing. (3) Remove the residual kernels on the corn cob after acquiring the corncob’s images. Then, measure the weight of the corncob. The ratio of the weight of the removed kernels to the weight of all kernels was the threshing rate of the corn ear, as shown in Equation (2).

$$\left\{\begin{array}{c}\rho =\frac{m}{M}\times 100\%\\ m=m_1-m_2\\ M=m_1-m_3\end{array}\right.$$

(2)

where $\rho$ is the threshing rate of the corn ear, $m$ is the weight of the removed kernels, $M$ is the weight of all kernels, $m_1$ is the weight of the corn ear before threshing, $m_2$ is the weight of the corncob with residual kernels after threshing, and $m_3$ is the weight of the corncob.

2.2. Measurement of Characteristic Parameters of the Corn Ear

The measurement of corn parameters mainly includes the length of the corncob, the diameter and radius of each segment, and the area of the top of all the corn kernels before threshing. The maximum range of vernier calipers used in the measurement is 300 mm, and the accuracy is 0.01 mm. (Because the area of the top of all the corn kernels in millimeters is relatively large, the units were converted to centimeters after measurement.)

2.2.1. Measurement of the Area of the Top of All the Corn Kernels

The unthreshed corn ear was equally divided into 30 segments by the principle of differentiation in mathematics (Figure 1a). Then, the diameter and radius of each ear were measured using vernier calipers. After differentiation, each segment of the corn ear in the middle part (part B in Figure 1b) is a cylinder, and both ends (parts A and C in Figure 1b) are a circular truncated cone.

The steps for measuring the area of the top of all corn kernels are as follows: (1) The area of the top of corn kernels in part B is calculated by Equation (3):

$$S_B={\sum }_i^n\frac{\pi L{r}_i}{15}$$

(3)

where $S_B$ is the area of the top of corn kernels in part B, $r_i$ is the radius of the underside of each cylindrical corn ear, and $L$ is the length of the whole corn ear.

(2) The area of the top of corn kernels in parts A and C is calculated by Equation (4):

$$S_{A+C}={\sum }_i^n\pi \left(r_i+r_i^{\prime }\right)\sqrt{{\left(r_i^{\prime }-r_i\right)}^2+{\left(\frac{L}{30}\right)}^2}$$

(4)

where $S_{A+C}$ is the area of the top of corn kernels in parts A and C, $r_i$ is the radius of the upper bottom surface of each circular truncated cone, $r_i^{\prime }$ is the radius of the lower bottom surface of each circular truncated cone, and $L$ is the length of the whole corn ear.

(3) Add the areas of the three parts A, B, and C to obtain the area of the top of all corn kernels, as shown in Equation (5):

$$S=S_{A+C}+S_B$$

(5)

where $S$ is the area of the top of all corn kernels, $S_{A+C}$ is the area of the top of corn kernels in parts A and C, and $S_B$ is the area of the top of corn kernels in part B.

2.2.2. Measurement of Corncob Parameters

The corncob is divided into three parts: the small section, midsection, and large section. Then, the length of the corncob and the diameter and radius of the small section, midsection, and large section of the corncob were measured using vernier calipers (Figure 2).

The radius of the small section, midsection, and large section were combined with the length of the corncob as independent variables, and the area of the top of all corn kernels was used as the dependent variable for multivariate linear fitting.

Table 1. The fitting degree between parameters of the corncob and the area of the top of all corn kernels.

Independent Variable	Dependent Variable	R2	p
Small-section radius, Length	Area of the top of all corn kernels	0.948	<0.001
Midsection radius, Length		0.987	<0.001
Large-section radius, Length		0.960	<0.001

shows the results of the fitting degree. When the midsection radius and length of the corncob are used as independent variables, the fitted equation R² is 0.987, indicating the highest fitting degree. Therefore, the following part of this paper chooses the midsection radius and length as the main parameters of the corncob to study.

The measured midsection radius and length of 300 corncobs were input into SPSS statistical software (IBM SPSS Statistics 27.0, which was developed by IBM in the United States) as variables for normality analysis. The results are shown in

Table 2. Corncob parameters.

Statistical Magnitude	Length (cm)	Midsection Radius (cm)
Max.	19.854	1.576
Min.	13.091	1.314
Mean	16.539	1.4666
S.D.	1.4002	0.0527
p	0.515	0.235

. The p values of the midsection radius and length of the corncob are much higher than 0.05. Therefore, they follow a normal distribution and can be used to build a bivariate linear regression area model (Figure 3).

2.3. The Method of Establishing the Regression Model

With the midsection radius and length of the corncob as independent variables and the area of all corn kernels as the dependent variable, linear regression analysis was carried out to establish a regression model to restore the area of all corn kernels. The steps of equation establishment are as follows: (1) The autocorrelation of independent variables is tested according to the variance inflation factor VIF. If there is a strong correlation between independent variables, it will cause a collinearity problem, which has an impact on regression. The test criteria are as follows: when 0 < VIF < 10, there is no multicollinearity; when 10 ≤ VIF < 100, there is strong multicollinearity; when VIF ≥ 100, there is severe multicollinearity. (2) Residual independence, residual normality, and residual variance homogeneity are sequentially examined. (3) The value of F in the results of ANOVA is used to analyze whether the linear relationship between the dependent variable and all independent variables in the model is significant in general. If F satisfies Equation (6), it is considered that the independent variables included in the model have a significant impact on the dependent variable; otherwise, there is no significant impact.

$$F>F_a(k,n-k-1)$$

(6)

where k is the number of independent variables, n is the sample size, and n − k − 1 is the degree of freedom.

2.4. Acquisition and Processing of Corn Images

Figure 4 shows the system for capturing images of corncobs, which mainly includes a computer, obscure, camera, light source, and carrier plate. The camera is equipped with an IMX703 sensor (Developed by SONY Corporation of Japan) featuring a 12-megapixel resolution and 1.9-micron pixel size, positioned 45 cm vertically from the carrier. Under the condition of sufficient light source, the corn is placed in the same position for shooting, and a complete picture of the front and back of the corncob is taken. The size of the captured image is 3024 × 4032 pixels.

Before image acquisition, nine calibration units (35 × 35 mm in size) were used to calibrate the image in the center position of the support plate according to the size of the corn (Figure 5). The result is that the width of the unit pixel is 0.054 mm. Therefore, the scale factor K of the image collected in this paper is set to 0.054 mm/pixel. The captured images were processed using python3.12 (Developed by the Python Software Foundation (PSF) in the Netherlands) and pycharm2023 (Developed by JetBrains in the Czech Republic) image processing software.

A threshold segmentation method based on the R-I space graph is used to binarize the image. The processing steps are as follows: (1) Extract the red channel (R) of the image. The brightness of the image (I) is calculated as the average of the gray values of the image. Then, the brightness is subtracted from the red channel to generate the R-I space diagram (Figure 6a). Its coordinates are composed of area dimension and instance dimension. Such a spatial map can intuitively represent the distribution of different regions and objects in the image. (2) Select the appropriate threshold value to segment the image according to the requirements through the R-I spatial map. Because this paper wants to segment corn kernels, corncob, and background, two thresholds of 10 and 40 are used to segment the image. The corn kernels and background are shown as black, and the corncob is shown as white to obtain a binary image (Figure 6b).

In order to improve the accuracy of corncob parameter detection, morphological treatment was used to eliminate the holes caused by different colors of corn kernels. The steps are as follows: (1) Small objects and noise in the image are eliminated by open operation. Open operation is the processing of an image successively through corrosion calculation and expansion calculation (Equation (7)). After the origin of the structure element is aligned with the origin of the original binary image’s upper left corner, the pixel’s gray value after the calculation can be obtained using the calculation relationship between the pixel and the neighboring pixel. (2) Small and medium-sized holes and cracks in the image are filled by closing operation. Closed operations are processed in reverse order to open operations. The process is that the image is first expanded and then corroded (Equation (7)).

$$\left\{\begin{array}{c}dst\left(x,y\right)=\underset{\left(x^{\prime },y^{\prime }\right):element\left(x^{\prime },y^{\prime }\right)\ne 0}{max}src\left(x+x^{\prime },y+y^{\prime }\right)\\ dst\left(x,y\right)=\underset{\left(x^{\prime },y^{\prime }\right):element\left(x^{\prime },y^{\prime }\right)\ne 0}{min}src\left(x+x^{\prime },y+y^{\prime }\right)\end{array}\right.$$

(7)

where element is the structural element, $\left(x,y\right)$ is the position of the origin, $\left(x^{\prime },y^{\prime }\right)$ is the position deviation of the structural element relative to the origin, $src$ is the original image, and $dst$ is the image after the operation.

A new binary image is obtained by open and close operations (Figure 6c). Small objects and noise are eliminated, and small voids and cracks are filled, keeping the shape and size of the corncob in the image unchanged.

2.5. Automatic Measurement Method of Corncob Parameters

The detection of corncob parameters is carried out according to the gray value feature of the binary image after processing (the gray value of the corncob region is 255) [22]. The detection method is shown in Figure 7. Its steps are as follows: (1) Convert the binarization image into a matrix of 3024 × 4032 and establish a coordinate system with the 4032nd line of the matrix as the X-axis and the first column as the Y-axis. (2) Extract the image coordinate information of all gray values of 255 and determine the center line $y_0$ in the direction of the Y-axis according to the maximum and minimum values of the Y coordinate. Scan the gray value array along the line $y_0$ and record the coordinate information with the gray value of 255 at the most ends, $X_1\left(x_1,y_0\right){,X}_1\left(x_2,y_0\right)$. The midsection radius $R$ of corncob is calculated by Equation (8):

$$\left\{\begin{array}{c}d=\left(x_2-x_1\right)\times K\\ R=\frac{1}{2}D=\frac{d_1+d_2}{4}\end{array}\right.$$

(8)

where $R$ is the midsection radius of the corncob, $d$ is the midsection diameter of the corncob, $d_1$ and $d_2$ are the midsection diameter of the same corncob in the front and back images, and $K$ = 0.054 mm/pixel.

(3) The length of the corncob from the tip of the small section to the end of the large section is different from the length of the area where the kernel is grown. Corncobs with a more pronounced tip have a slightly shorter length in the area of the kernel growth. In order to reduce the influence of the tip, the average length of the center part is used to determine the length of the corncob. The center line $x_0$ in the direction of the X-axis is determined according to the two horizontal coordinates $x_1$ and $x_2$. Scan an array of grayscale values along line $x_0$ for five columns to its left and right. Record the coordinate information $M_i\left(x_i,y_i\right)$ and $N_i\left(x_i,y_i^{\prime }\right)$ of each column’s highest and lowest points with the gray value of 255. The length H of corncob is calculated by Equation (9):

$$\left\{\begin{array}{c}h=\frac{{\sum }_i^n\left(y_i-y_i^{\prime }\right)}{n}\\ H=\frac{h_1+h_2}{2}\times K\end{array}\right.$$

(9)

where $H$ is the length of the corncob, $h$ is the length of the corncob in the image, $h_1$ and $h_2$ are the length of the same corncob in the front and back images, and $K$ = 0.054 mm/pixel.

3. Results and Discussion

3.1. The Result of Establishing the Regression Model

Table 3. Coefficient model.

	Unstandardized Coefficients		Beta	t	p	VIF
	B	Std. Error
Constant	−220.976	12.602		17.535	<0.0001
Length	14.567	0.310	0.845	47.061	<0.0001	1.164
Midsection radius	150.946	9.195	0.295	16.416	<0.0001	1.164

shows the coefficient model of the regression equation. The p of the corncob’s length and midsection radius is less than 0.0001. Moreover, their corresponding VIF values are between 0 and 10. The results showed that the length and midsection radius of the corncob did not have multicollinearity, and the influence on the area of the top of all corn kernels was considerable.

The model summary information is presented in

Table 4. Summary information of the model.

R	R2	Adjusted R2	D-W
0.993a	0.987	0.986	2.053

. The coefficient of determination (R²) indicates the goodness of fit, measuring how well the estimated model fits the dependent variable. A higher R² value closer to 1 suggests a better-fitting model. In this case, R² is calculated as 0.987, and the Durbin–Watson statistic (DW) is 2.053, indicating a solid fit and residual independence.

Figure 8 is a normal P-P plot of the regression-standardized residuals to test the normality of the residuals. The underlying principle is that if the data follow a normal distribution, their cumulative proportion should align with a standard normal distribution. The actual data accumulation ratio is taken as the X-axis, and the corresponding normal distribution accumulation ratio is taken as the Y-axis to make a scatter plot. Most of the points fall near and fluctuate around the directly proportional function line, which proves that the residual is the normal distribution.

The scatter plot in Figure 9 is used to assess the homogeneity of the residual variance through standardized residuals. The criterion for this test is that each residual should be randomly distributed between −2 and 2 on the scatter plot, showing no discernible pattern. Any predictable information within the residuals should not be present. Heteroscedasticity can be indicated by a regular appearance in the scatter plot, such as a linear or trumpet shape. Most of the scattered points in the diagram fluctuate around the ordinate 0. The points are distributed randomly and evenly between −2 and 2, with occasional deviations from the data. Moreover, there is no consistent pattern. These cases show that the residual variance is homogeneous.

Table 5. ANOVA model.

	Sum of Squares	df	Mean Square	F	p
Regression	24,803.425	2	12,401.712	1783.147	<0.001
Residuals	326.883	47	6.955
Total	25,130.308	49

is the ANOVA model, representing the analysis of variance results. The result shows that p is less than 0.001, and the $F_a$ can be calculated as 3.0258. F in the table is significantly greater than $F_a$, proving that the independent variables in the model significantly affect the dependent variable, and the linear regression equation obtained is good.

In summary, taking the middle radius and length of the corncob as independent variables, the binary linear regression model of the area of the top of all corn kernels was established as follows:

$$\mathrm{S}=150.946\times \mathrm{R}+14.567\times \mathrm{H}-220.976$$

(10)

where S is the area of the top of all corn kernels, R is the midsection radius of the corncob, and H is the length of the corncob.

3.2. The Results of Image Processing and Corncob Parameters’ Automatic Measurement

Figure 10 shows the results of each step of image processing. In the binarized figure, glume shells were identified as corn kernels. The misjudged parts were removed by morphological operation and hole filling, and large pores were filled. The final region of corn kernels is basically consistent with the original image.

The parameters of 70 corncobs were automatically measured. Figure 11 shows a case of corncob parameters’ automatic measurement.

Table 6. The results of corncob parameters’ automatic measurement.

Statistical Magnitude	Length (cm)	Midsection Radius (cm)
Mean	16.136	1.307
Max. relative error	7.46%	5.55%
Min. relative error	0.05%	0.03%
Mean relative error	2.58%	2.32%

shows the automatic measurement results of the corncob’s length and midsection radius.

Figure 12 is a scatter plot of the relative error statistics of the automatic measurement of the corncob’s length and midsection radius. The maximum relative error of the length of the automatic measurement is 7.46%, and the mean relative error is 2.58%. The maximum relative error of the midsection radius is 5.55%, and the mean relative error is 2.23%. The mean relative error of the corncob’s midsection radius measurement was better than that from Hu et al. [31], who measured lettuce height using Kinect (2.58%).

Figure 13 is a scatterplot with manually measured values as horizontal coordinates and automatically measured values as vertical coordinates. The points in the graph are distributed near the proportional function. The two lines in the figure are the linear regression analysis fitting lines of the automatic and manual measurements of the radius and length of the middle part of the corncob. The regression equation R² of the radius of the middle section is 0.93. The regression equation R² of the length is 0.89. Their fitting effect is better, which indicates that the automatic measurement value is close to the true value. Combining Figure 11 and Figure 12, the results of the automatic measurement of the midsection radius are superior to those of the automatic measurement of length. This phenomenon is mainly caused by the location of image acquisition. Since the camera is located directly above the middle part of the corncob, the pixel distortion in the middle position of the image is minor, and the imaging error is correspondingly minor.

3.3. The Results of Detection of the Threshing Rate of Corn Ears

The automatic measurement of the threshing rate is achieved by substituting the automatic measurement results of the corncob’s parameters into Formulas (1) and (10). The automatic measurement values of the threshing rate of 70 corncobs were compared with the manual weighing measurement values, as shown in Figure 14. The coincidence between the automatic measurement curve and the manual measurement curve is significant. The maximum relative error of automatic measurement is 7.08%, and the mean relative error is 2.04%. This indicates that the automatic detection results are close to the actual value. However, both the curve and scatter points in the figure have large deviations. In order to explore the reason for this situation, the corncob was classified for a comparative experiment.

The coincidence between the automatic measurement curve and the manual measurement curve is significant. The maximum relative error of automatic measurement is 7.08%, and the mean relative error is 2.04%. This indicates that the automatic detection results are close to the actual value. However, both the curve and scatter points in the figure have large deviations. In order to explore the reason for this situation, the corncob was classified for a comparative experiment.

The distribution of residual kernels on the corncob is divided into three categories: large-section concentrated, midsection concentrated, and small-section concentrated (hereinafter referred to as large section, midsection, and small section). We selected 20 corncobs of each category for the experiment. The results are shown in

Table 7. The results of the threshing rate of corn ears detected by different categories of corncobs.

Corncob	The Weighing Method			Image Processing			Error
	Large	Mid	Small	Large	Mid	Small	Large	Mid	Small
1	92.94%	97.39%	97.39%	95.36%	97.37%	96.39%	2.42%	0.02%	0.10%
2	93.88%	95.27%	95.27%	96.37%	94.96%	96.23%	2.49%	0.31%	1.59%
3	93.01%	97.02%	97.02%	94.78%	95.90%	95.26%	1.77%	1.12%	1.34%
4	98.27%	89.23%	89.23%	94.43%	92.47%	96.13%	3.84%	3.24%	4.67%
5	89.40%	87.00%	87.00%	91.73%	90.61%	93.65%	2.33%	3.61%	5.84%
6	91.77%	92.47%	92.47%	93.33%	93.00%	94.77%	1.56%	0.53%	0.17%
7	79.65%	96.51%	96.51%	85.72%	97.50%	97.02%	6.07%	0.99%	1.82%
8	92.75%	91.37%	91.37%	95.84%	92.60%	98.00%	3.09%	1.23%	2.79%
9	82.43%	88.05%	88.05%	80.12%	88.46%	90.88%	2.31%	0.41%	1.85%
10	83.18%	90.32%	90.32%	79.27%	90.86%	90.85%	3.91%	0.54%	3.27%
11	90.16%	92.17%	92.17%	87.47%	93.13%	97.45%	2.69%	0.96%	3.83%
12	90.93%	88.14%	88.14%	88.62%	88.50%	97.36%	2.31%	0.36%	2.78%
13	95.58%	97.05%	97.05%	97.56%	98.00%	93.97%	1.98%	0.95%	1.83%
14	95.25%	91.39%	91.39%	97.71%	92.22%	96.65%	2.46%	0.83%	2.09%
15	95.44%	96.23%	96.23%	97.91%	97.28%	94.51%	2.47%	1.05%	2.13%
16	94.75%	92.48%	92.48%	97.99%	92.21%	89.09%	3.24%	0.27%	2.12%
17	89.92%	92.97%	92.97%	92.32%	93.88%	85.82%	2.40%	0.91%	2.55%
18	94.64%	88.48%	88.48%	97.82%	87.94%	85.88%	3.18%	0.54%	4.41%
19	89.97%	91.16%	91.16%	88.26%	92.16%	98.92%	1.71%	1.00%	1.48%
20	90.66%	91.78%	91.78%	89.08%	92.81%	92.76%	1.58%	1.03%	1.94%

.

Figure 15 shows the statistical results of relative errors in the automatic measurement of the corn ear threshing rate under three categories. Combining Table 7 and Figure 15, the maximum relative errors of the large section, midsection, and small section are 7.08%, 3.98%, and 6.24%, respectively. The mean relative errors are 2.95%, 1.07%, and 2.60%, respectively. Automatic measurement results are better when residual kernels are concentrated in the middle section. This is because the midsection of the corncob is relatively flat, and there is a particular slope at both ends. When residual kernels are concentrated in the midsection, the automatic measurement of the area of the top of the residual kernels is more accurate.

4. Conclusions

In order to realize the rapid measurement of corn ears’ threshing rate, this paper uses the ratio of the area of the top of the missing kernels after threshing to the area of the top of all kernels to calculate the threshing rate of corn ears based on machine vision technology. The mathematical relationship between the diameter, radius, and length of the corncob and the area of the top of all kernels was analyzed to restore the area of the top of all corn kernels. The corncobs’ parameters and the area of the top of the residual kernels were measured by machine vision to detect the threshing rate of corn ears. The results are as follows:

(1)

The regression model for restoring the area of the top of all corn kernels of an ear by midsection radius and length of the corncob was established. The regression equation with the R² value exceeding 0.98 demonstrates a strong level of fit.

(2)

A method of measuring corncobs’ parameters by image processing is proposed. The maximum relative error of length and midsection radius was 7.46% and 5.55%, and the mean relative error was 2.58% and 2.23%.

(3)

A method based on machine vision to detect the threshing rate of corn ears by the area of the top of corn kernels was proposed. Compared with the weighing method, the maximum relative error of automatic measurement is 7.08%, and the mean relative error is 2.04%. When the residual kernels were concentrated in the midsection, the inspection result of the corn ear threshing rate was better. The maximum relative error was 3.98%, and the mean relative error was 1.07%.

There are still some issues that require further study. More experiments in terms of different corn varieties need to be conducted to improve the performance of the proposed method. Additionally, according to the threshing rate of each corn ear, an automatic sorting device of corncobs after threshing can be designed, thereby enhancing raw material utilization and reducing costs.