CN116452526A - A Rice Seed Recognition and Counting Method Based on Image Detection - Google Patents

Abstract

The invention provides a rice seed identification and counting method based on image detection, which adopts grid paper as a background for rice seed image acquisition, thereby realizing positioning and correction of images; then selecting color components with obvious differences between rice seeds and the background to form an SbCBCr color space, and dividing the rice seed image into binary images by combining a BPNN classification model; after the open-close operation, dividing the original image into sub-images of single rice seeds or a plurality of adhered rice seeds by taking the binary image communication area as a mask, and carrying out size normalization; and establishing a rice seed number image data set by the sub-images, classifying the acquired rice seed images by using a network classification model by using a deep convolution god, and then counting the rice seed number corresponding to each sub-image classification result in the original rice seed image to obtain the total number of rice seeds in the original rice seed image. The invention has simple operation and accurate counting, can save labor cost and equipment cost and greatly improve the working efficiency.

Description

A Rice Seed Recognition and Counting Method Based on Image Detection

technical field

The invention belongs to the field of image processing, and in particular relates to a method for identifying and counting rice seeds based on image detection.

Background technique

Accurately counting the number of seeds is an indispensable step in modern agricultural scientific research, and an important link in breeding and testing. Crop seed research mainly lies in quality detection and counting, and the identification and segmentation of seeds are the most important links in the counting of crop seeds. Most of the research focuses on the segmentation of sticky crop seeds, and the segmentation algorithm based on morphology, pit detection and elliptic curve fitting is the basic method. There are other algorithms like segmentation algorithm based on seed region growing, segmentation algorithm based on clustering algorithm and watershed algorithm etc. However, the current counting method has higher requirements on the environment and pictures, and the speed is relatively slow, so its convenience, accuracy, and quickness need to be improved.

At present, the method of manual counting is still very common. The advantages are high real-time performance and low tool requirements, while the disadvantages are low efficiency, error-prone, labor-intensive, and visual fatigue. There are also a small number of electromechanical integrated counting equipment that can replace manual counting, but there are problems such as large errors, complicated manufacturing, and high prices, making it difficult to widely popularize and apply. To sum up, in order to solve the above-mentioned deficiencies in the prior art, the present invention provides an image detection-based rice seed identification and counting method on the basis of image processing technology.

Contents of the invention

Purpose of the invention: Aiming at the deficiencies in the prior art, the present invention proposes a method for identifying and counting rice seeds based on image detection, and uses a deep convolutional neural network classification model to realize the counting of rice seed adhesions, achieving fast speed , high accuracy and strong stability.

Technical solution: The present invention provides a method for identifying and counting rice seeds based on image detection, which specifically includes the following steps:

(1) Sprinkle the rice seeds evenly on the white grid paper, the camera lens is parallel to the top of the grid paper for overhead shooting, and a parallel light source is used for supplementary light;

(2) Rotate and crop the image according to the position of the four corner positioning blocks on the grid paper, and then calibrate the image distortion according to the position of the grid intersection in the grid paper;

(3) Extract the S component of the rice seed image in the HSV color space, the b component of the Lab color space, and the Cb and Cr components of the YCbCr color space to form a new color space SbCbCr;

(4) Build the BPNN pixel classification model, then extract the pixel values of the four channels in the new color space SbCbCr of the seed pixel and the background pixel in the sample image, train the BPNN model, and select the optimal model;

(5) The color value of each pixel in the rice seed image in the four-channel SbCbCr color space is used as the input of the optimal BPNN model, and the rice seed image is divided into binary images, wherein the rice seed is divided into foreground and grid paper is segmented into the background;

(6) Use the open operation to disconnect the small adhesions in the connected areas of some rice seeds and filter out the tiny noise points, and then use the closed operation to make the edges of the rice seeds smoother;

(7) With each rice seed connected region as a mask, the original rice seed image is divided into rice seed images composed of a single rice seed or a plurality of sticky rice seeds, and then size normalized to it;

(8) Label according to the number of rice seeds in each sub-graph, so as to establish the image data set of rice seeds, use the depth convolution algorithm to establish an image classification model with the network, and input the image data set into the classification model for training;

(9) Divide the collected rice seed image into multiple sub-pictures and send them into the trained classification model, then count the number of rice seeds corresponding to each rice seed image classification result in the original rice seed image, and obtain the rice seed in the original rice seed image the total number of .

Further, the implementation process of the step (2) is as follows:

(21) according to the color of the grid paper positioning block, the positioning block is separated from the image;

(22) adopt corner detection method to determine the vertex position of each positioning block;

(23) connect the vertices of four positioning blocks, and calculate the deflection angle of the placement of grid paper;

(24) Perform reverse rotation and cropping operations on the collected image according to the deflection angle, so as to straighten the image;

(25) Use corner detection to determine the position of each intersection point in the grid, by calculating the horizontal and vertical distances of adjacent intersection points in different positions in the image, and subtracting it from the actual side length of the grid in the image, thereby Calculate the distortion rate of the image in the horizontal and vertical directions;

(26) Perform image calibration according to the distortion rate.

Further, the implementation process of the step (3) is as follows:

The calculation methods of different color components, the calculation formula of S component in HSV color space is shown in formula (1), the calculation formula of b component in Lab color space is shown in formula (2), and the Cb and Cr components in YCbCr color space The calculation formulas of are shown in formula (4);

b=200(h(Y/Y _w )-h(Z/Z _w )) (2)

In the formula, MAX and MIN are the maximum and minimum values of the three color components in the RGB color space, respectively; Y and Z are the corresponding components in the XYZ color space, respectively, and the reference values of Y _w and Z _w are 1.0000 and 1.0888, respectively, The color stimulus value calibration function h(t) is shown in formula (3); R, G, and B are three components of the RGB color space, respectively.

Further, the implementation process of the step (4) is as follows:

(41) The number of nodes of the BPNN model input layer is 4, and the four color channel components of each pixel in the SbCbCr color space are used as input data;

(42) The number of nodes in the output layer of the BPNN model is 1, and the rice seed pixel sample is marked as 1 as a positive sample, and the background pixel sample is marked as 0 as a negative sample;

(43) The quantity of BPNN model hidden layer node is determined as 10, the activation function of setting hidden layer is Sigmoid, the activation function of output layer is Softmax, and adopts cross entropy loss function to measure prediction error;

(44) Use the quantized conjugate gradient function for multiple trainings, and calculate and analyze parameters such as loss, error, accuracy, true rate and false positive rate of different training models, so as to select the optimal BPNN model.

Further, the implementation process of the step (5) is as follows:

(51) convert the rice seed image of M*N into a matrix of (M*N, 4), each row of data in the matrix corresponds to a pixel of the image, and each row of data corresponds to a color component of the SbCbCr color space;

(52) Carry out normalization processing to matrix data, and input optimal BPNN model;

(53) Quantize the M*N-dimensional column vector output by the model, set it to 1 if it is greater than or equal to 0.5, and set it to 0 if it is less than 0.5;

(54) The column vector is converted into a binary image of M*N size, the pixel point with a value of 1 is the foreground to represent the rice seed, and the pixel point with a value of 0 is the background to represent the grid paper.

Further, the implementation process of the step (7) is as follows:

(71) Calculate the coordinates of the pixel points at the top, bottom, leftmost and rightmost of each connected region;

(72) Draw an external rectangle according to the pixel coordinates in four directions;

(73) mapping the circumscribed rectangle in the binary image to the original image, and cutting out a subgraph containing a single rice seed or a plurality of glued rice seeds in the corresponding area of the original image;

(74) Normalize the size of the cropped sub-images.

Beneficial effect: Compared with the prior art, the beneficial effect of the present invention is that the present invention applies digital image processing technology to seed counting in the field of breeding, replacing manual counting and mechanical counting, saving labor costs and equipment costs, and substantially Improve work efficiency; the present invention uses grid paper with positioning blocks to accurately identify and segment rice seeds; the present invention uses a deep convolutional neural network classification model to realize the counting of seed adhesions, achieving fast speed and high accuracy , strong stability, batch processing and other advantages; the invention is easy to operate, low in cost, and can realize accurate prediction of the number of rice seeds.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is the flowchart of establishing BPNN pixel classification model;

Fig. 3 is the binary image of rice seed;

Fig. 4 is an image of rice seed morphology processing.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

The invention proposes a method for identifying and counting rice seeds based on image detection. The rice seeds are evenly sprinkled on the white grid paper, the camera lens is parallel to the top of the grid paper for overhead shooting, and a parallel light source is used for supplementary light. Import the collected image into the computer, rotate and crop the image according to the position of the four corner positioning blocks of the grid paper, and then calibrate the image distortion according to the position of the intersection point of the grid in the grid paper; extract the S component of the rice seed image in the HSV color space, The b component of the Lab color space and the Cb and Cr components of the YCbCr color space are combined into a new color space SbCbCr. Establish a BPNN pixel classification model, then extract the pixel values of the four channels of the seed pixel and the background pixel in the sample image in the new color space SbCbCr, perform parameter training on the BPNN model, and select the optimal model; The color values of the four channels in the SbCbCr color space are used as the input of the optimal BPNN model, and the rice seed image is segmented into a binary image, in which the rice seed is segmented as the foreground, and the grid paper is segmented as the background. The opening operation is used to disconnect the small adhesions in the connected areas of some rice seeds and to filter out tiny noise points, and then the closing operation is used to make the edges of the rice seeds smoother. Using the connected area of each rice seed as a mask, the original rice seed image is divided into rice seed images composed of a single rice seed or multiple contiguous rice seeds, and then the size is normalized; according to the The number of rice seeds is labeled, so as to establish an image dataset of the number of rice seeds, and a deep convolution network is used to establish an image classification model, and the image data set is input into the classification model for training. Divide the collected rice seed images into multiple sub-images and send them to the trained classification model, then count the number of rice seeds corresponding to each rice seed image classification result in the original rice seed image, and obtain the total number of original rice seed images . As shown in Figure 1, it specifically includes the following steps:

Step 1: Sprinkle the rice seeds evenly on the white grid paper. The camera lens is parallel to the top of the grid paper for overhead shooting, and a parallel light source is used to fill in the light, so that the collected images are clear and bright, and the influence of shadows is reduced. Use grid paper with positioning blocks as the background for shooting, and use parallel light sources to reduce shadows.

Step 2: Rotate and crop the image according to the positions of the four corner positioning blocks on the grid paper, and then calibrate the image distortion according to the positions of the intersection points of the grid on the grid paper.

According to the color of the grid paper positioning block, separate the positioning block from the image; use the corner detection method to determine the vertex position of each positioning block; connect the vertices of the four positioning blocks, and calculate the deflection of the grid paper placement Angle: According to the deflection angle, reverse rotation and cropping operations are performed on the collected image to straighten the image. Use corner detection to determine the position of each intersection point in the grid, and then calculate the distance between horizontal and vertical adjacent corner points, and subtract it from the actual side length of the grid in the image, so as to calculate the image in the grid. The distortion rate in the horizontal and vertical directions, and finally perform image calibration according to the distortion rate.

Step 3: Extract the S component of the rice seed image in the HSV color space, the b component of the Lab color space, and the Cb and Cr components of the YCbCr color space to form a new color space SbCbCr.

The calculation methods of different color components, the calculation formula of S component in HSV color space is shown in formula (1), the calculation formula of b component in Lab color space is shown in formula (2), and the Cb and Cr components in YCbCr color space The calculation formulas of are shown in formula (4).

b=200(h(Y/Y _w )-h(Z/Z _w )) (2)

In the formula, MAX and MIN are the maximum and minimum values of the three color components in the RGB color space, respectively; Y and Z are the corresponding components in the XYZ color space, respectively, and the reference values of Y _w and Z _w are 1.0000 and 1.0888, respectively, Among them, the function color stimulus value calibration function h(t) is shown in formula 3; R, G, B are three components of RGB color space respectively.

Step 4: Establish a BPNN pixel classification model, then extract the pixel values of the four channels of the seed pixel and the background pixel in the sample image in the new color space SbCbCr, perform parameter training on the BPNN model, and select the optimal model.

The details of training the BPNN pixel classification model are shown in Figure 2, including:

S1: The number of nodes in the input layer of the BPNN model is 4, and the four color channel components of each pixel in the SbCbCr color space are used as input data.

S2: The number of nodes in the output layer of the BPNN model is 1, the rice seed pixel sample is marked as 1 as a positive sample, and the background pixel sample is marked as 0 as a negative sample.

S3: Set the number of nodes in the hidden layer of the BPNN model to 10, set the activation function of the hidden layer to Sigmoid, and the activation function of the output layer to Softmax, and use the cross-entropy loss function to measure the prediction error.

S4: Use the quantized conjugate gradient function to conduct multiple trainings, and calculate and analyze the parameters of different training models such as loss, error, accuracy rate, true rate and false positive rate, so as to select the optimal BPNN model.

Step 5: The color value of each pixel in the rice seed image in the four channels in the SbCbCr color space is used as the input of the optimal BPNN model, and the rice seed image is segmented into a binary image, as shown in Figure 3, where the rice seed is segmented For the foreground, the grid paper is split for the background.

Convert the M*N rice seed image into a matrix of (M*N, 4), each row of data in the matrix corresponds to a pixel of the image, and each column of data corresponds to a color component of the SbCbCr color space. Normalize the matrix data and input it into the optimal BPNN model. Quantize the M*N-dimensional column vector output by the model, set it to 1 if it is greater than or equal to 0.5, and set it to 0 if it is less than 0.5. Convert the column vector into a binary image of M*N size. In Figure 3, the pixel with a value of 1 is the foreground to represent the rice seed, and the pixel with the value of 0 in the figure is the background to represent the grid paper.

Step 6: Use the open operation to disconnect the small adhesions in the connected areas of some rice seeds and filter out tiny noise points, and then use the close operation to make the edges of the rice seeds smoother. A square structural element with a side length of 2 is selected for opening operation processing, and a square structural element with a side length of 2 is selected for closing operation processing; the image after the opening and closing operation is shown in Figure 4.

Step 7: Using the connected area of each rice seed as a mask, the original rice seed image is divided into rice seed images composed of a single rice seed or a plurality of conglutinated rice seeds, and then the size is normalized.

Calculate the pixel coordinates of the uppermost, lowermost, leftmost and rightmost of each connected region; draw the outer rectangle according to the pixel coordinates in the four directions. Map the circumscribed rectangle in the binary image to the original image, and cut out a sub-image containing a single rice seed or multiple cohesive rice seeds in the corresponding area of the original image; normalize the size of the cropped sub-image.

Step 8: Label according to the number of rice seeds in each sub-graph, so as to establish an image data set of rice seeds, use the deep convolution neural network to establish an image classification model, and input the image data set into the classification model for training.

Label according to the number of rice seeds in each sub-graph, and establish an image dataset of rice seeds; establish a deep convolutional neural network classification model, the input of the model is consistent with the size of the dataset image, and the output is the number of rice seeds; the dataset image Send the deep convolutional neural network model for multiple trainings, test the convolutional neural network model, and obtain the trained convolutional neural network model.

Step 9: Divide the collected rice seed images into multiple sub-images and send them to the trained classification model, then count the number of rice seeds corresponding to the classification results of each rice seed image in the original rice seed image, and obtain the number of rice seeds in the original rice seed image The total number of.

Complete the collection of rice seed images, and divide the collected images into size-normalized subgraphs containing single or multiple sticky rice seeds; send the normalized subgraphs to the trained image classification model for classification; count each The classification results of each sub-map are summed to obtain the exact number of rice species in the collection map.

The method is accurate in calculation and simple in operation, can effectively count rice seeds accurately, saves labor costs and equipment costs, and greatly improves work efficiency.

It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and not to limit the present invention. Therefore, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present invention shall fall within the protection scope of the present invention. Furthermore, it is intended that the appended claims of the present invention embrace all changes and modifications that come within the scope and metesques of the appended claims, or equivalents of such scope and metes and bounds.

Claims (6)

1. a rice seed identification and counting method based on image detection, is characterized in that, comprises the following steps: (1) Sprinkle the rice seeds evenly on the white grid paper, the camera lens is parallel to the top of the grid paper for overhead shooting, and a parallel light source is used for supplementary light; (2) Rotate and crop the image according to the position of the four corner positioning blocks on the grid paper, and then calibrate the image distortion according to the position of the grid intersection in the grid paper; (3) Extract the S component of the rice seed image in the HSV color space, the b component of the Lab color space, and the Cb and Cr components of the YCbCr color space to form a new color space SbCbCr; (4) Build the BPNN pixel classification model, then extract the pixel values of the four channels in the new color space SbCbCr of the seed pixel and the background pixel in the sample image, train the BPNN model, and select the optimal model; (5) The color value of each pixel in the rice seed image in the four-channel SbCbCr color space is used as the input of the optimal BPNN model, and the rice seed image is divided into binary images, wherein the rice seed is divided into foreground and grid paper is segmented into the background; (6) Use the open operation to disconnect the small adhesions in the connected areas of some rice seeds and filter out the tiny noise points, and then use the closed operation to make the edges of the rice seeds smoother; (7) With each rice seed connected region as a mask, the original rice seed image is divided into rice seed images composed of a single rice seed or a plurality of sticky rice seeds, and then size normalized to it; (8) Label according to the number of rice seeds in each sub-graph, so as to establish the image data set of rice seeds, use the depth convolution algorithm to establish an image classification model with the network, and input the image data set into the classification model for training; (9) Divide the collected rice seed image into multiple sub-pictures and send them into the trained classification model, then count the number of rice seeds corresponding to each rice seed image classification result in the original rice seed image, and obtain the rice seed in the original rice seed image the total number of . 2. a kind of rice seed identification and counting method based on image detection according to claim 1, is characterized in that, described step (2) realization process is as follows: (21) according to the color of the grid paper positioning block, the positioning block is separated from the image; (22) adopt corner detection method to determine the vertex position of each positioning block; (23) connect the vertices of four positioning blocks, and calculate the deflection angle of the placement of grid paper; (24) Perform reverse rotation and cropping operations on the collected image according to the deflection angle, so as to straighten the image; (25) Use corner detection to determine the position of each intersection point in the grid, by calculating the horizontal and vertical distances of adjacent intersection points in different positions in the image, and subtracting it from the actual side length of the grid in the image, thereby Calculate the distortion rate of the image in the horizontal and vertical directions; (26) Perform image calibration according to the distortion rate. 3. a kind of rice seed identification and counting method based on image detection according to claim 1, is characterized in that, described step (3) realization process is as follows: The calculation methods of different color components, the calculation formula of S component in HSV color space is shown in formula (1), the calculation formula of b component in Lab color space is shown in formula (2), and the Cb and Cr components in YCbCr color space The calculation formulas of are shown in formula (4); b=200(h(YY _w )-h(ZZ _w ))(2) In the formula, MAX and MIN are the maximum and minimum values of the three color components in the RGB color space, respectively; Y and Z are the corresponding components in the XYZ color space, respectively, and the reference values of Y _w and Z _w are 1.0000 and 1.0888, respectively, The color stimulus value calibration function h(t) is shown in formula (3); R, G, and B are three components of the RGB color space, respectively. 4. a kind of rice seed identification and counting method based on image detection according to claim 1, is characterized in that, described step (4) realization process is as follows: (41) The number of nodes of the BPNN model input layer is 4, and the four color channel components of each pixel in the SbCbCr color space are used as input data; (42) The number of nodes in the output layer of the BPNN model is 1, and the rice seed pixel sample is marked as 1 as a positive sample, and the background pixel sample is marked as 0 as a negative sample; (43) The quantity of BPNN model hidden layer node is determined as 10, the activation function of setting hidden layer is Sigmoid, the activation function of output layer is Softmax, and adopts cross entropy loss function to measure prediction error; (44) Use the quantized conjugate gradient function for multiple trainings, and calculate and analyze parameters such as loss, error, accuracy, true rate and false positive rate of different training models, so as to select the optimal BPNN model. 5. a kind of rice seed identification and counting method based on image detection according to claim 1, is characterized in that, described step (5) realization process is as follows: (51) convert the rice seed image of M*N into a matrix of (M*N, 4), each row of data in the matrix corresponds to a pixel of the image, and each row of data corresponds to a color component of the SbCbCr color space; (52) Carry out normalization processing to matrix data, and input optimum BPNN model; (53) Quantize the M*N-dimensional column vector output by the model, set it to 1 if it is greater than or equal to 0.5, and set it to 0 if it is less than 0.5; (54) The column vector is converted into a binary image of M*N size, the pixel point with a value of 1 is the foreground to represent the rice seed, and the pixel point with a value of 0 is the background to represent the grid paper. 6. a kind of rice seed identification and counting method based on image detection according to claim 1, is characterized in that, described step (7) realization process is as follows: (71) Calculate the coordinates of the pixel points at the top, bottom, leftmost and rightmost of each connected region; (72) Draw an external rectangle according to the pixel coordinates in four directions; (73) mapping the circumscribed rectangle in the binary image to the original image, and cutting out a subgraph containing a single rice seed or a plurality of glued rice seeds in the corresponding area of the original image; (74) Normalize the size of the cropped sub-images.