JP2952340B2 - Crop detection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a crop detection method suitable for automation of agricultural work, and more particularly to a crop which extracts only a crop from image information in which a crop and other objects are mixed by a fuzzy inference method. The detection method.

[0002]

2. Description of the Related Art In order to automate farming operations such as thinning, weeding, harvesting, and the like, it is premised on establishing a technique for causing a working machine to recognize the position and size of a crop.

Research on crop recognition has been active in recent years,
In the case of an outdoor field, a visual sensor is often used since a large amount of information is required for discriminating the outside world from the target.

Various sensors have been developed as visual sensors, such as an RGB color camera for detecting a visible light region, and an electromagnetic wave of any wavelength from infrared to ultraviolet. In addition, many algorithms for extracting crops from image information obtained by a visual sensor include two-dimensional methods for extracting features such as color, brightness, shape, size, and the like to extract a target object in a planar manner. .

[0005]

However, none of the methods has improved the detection accuracy to a practical level.
The reason for this is that the detection characteristics of the sensor change due to changes in natural environmental conditions such as light and temperature, which affects the detection accuracy. In addition, the conditions of soil, weeds, and crops in the outdoor fields to be identified are not constant, and their colors, shapes, and sizes change every moment depending on the location and time. One of the major reasons is that a detection algorithm that does not have a sufficient capacity cannot be used.

Incidentally, the following three patterns can be raised as image processing error patterns. Eliminate crops Leave noise (substances other than crops) Eliminate crops and leave noise

[0007] For the purpose of agricultural work, it is not permissible to delete crops, so in order to reliably detect crops under severe environmental conditions, the conditions for discriminating between crops and others must be relaxed.
We must select an algorithm that is likely to cause the pattern.

As a method for alleviating such a restriction, a crop is extracted from an image containing noise by utilizing the fact that the arrangement of the planted crop, such as the distance and direction between individuals, is regular. It is possible. However, in an actual field, there is usually an error in regularity due to variations in planting intervals and the like.

By the way, most conventional image recognition uses a linear pattern matching method. However, the recognition rate is reduced unless the recognition target has a standardized pattern. In addition, when targeting an outdoor field, the measured values of the distance and direction between the binarized objects are expected to include a considerable amount of errors due to errors occurring in the complicated image processing process up to that time. That is, the measurement value including the error is
Although the arrangement is estimated in consideration of the regularity, the actual arrangement is not always according to the set rule.

Under such conditions, the application of a non-linear method is effective. That is, for example, it is thought that array estimation by pattern matching with a learning function provided to a work machine by a neural network is possible, but planting styles are diverse depending on the crops and local practices, and all patterns are learned. There are difficulties to make it happen.

The present invention has been made in view of the above circumstances, and has a fuzzy logic that is flexible in ambiguity and can easily cope with a change in the setting of an array or a change in the scale of an array handled in one screen. It is an object of the present invention to provide a crop detection method that uses the regularity of the arrangement of planted crops to discriminate noise on the screen from crops and enables only crops to be extracted. .

[0012]

In order to achieve the above object, the method for detecting a crop according to the present invention utilizes the regularity of the arrangement of planted crops to change the arrangement of inter-row and inter-strain arrangements, and to perform the method described above.
By using a fuzzy inference method that can easily respond to changes in the number of lines to be processed in a screen and the number of crops per line in a screen, image information in which crops and other objects are mixed can be used.
An object having the closest positional relationship to the set crop arrangement is regarded as a crop, and only the crop is extracted.

[0013]

The crop detection method of the present invention utilizes the regularity of the arrangement of the planted crops, is flexible in ambiguity, and can be easily changed in the arrangement setting or the scale of the arrangement handled in one screen. By using a fuzzy theory that can handle, from the image information where the crop and noise (substances other than the crop) are mixed, the object that has the closest positional relationship to the set crop arrangement is regarded as the crop. Extraction becomes possible.

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings. That is, in the present embodiment, an object having the closest positional relationship to the set crop arrangement is regarded as a crop and extracted from image information in which crops and noise (substances other than crops) are mixed. is there. The image to be processed as described below is R
This is a binary image output by any image processing method such as GB image processing.

Hereinafter, the method for detecting a crop according to the present embodiment will be described with reference to the procedure shown in FIG. First, after a binary image is output by image processing, the coordinates of the center of gravity of the object are calculated, and then the setting conditions are input (step 10).
1-103). That is, as shown in the schematic diagram of the setting arrangement in FIG. 2, it is premised that the crops 10 as objects are arranged in a lattice, and so-called staggered planting is not targeted.

As a set value representing the regularity of the arrangement, a set interval: L1 a set strain: L2 The number of rows to be processed in one screen: I The number of crop individuals per row in one screen: J is input.

However, the values of L1 and L2 are arbitrary. Also, I and J are two or more. Each crop 1 on the screen
The position of 0 is represented by the XY coordinates of each center of gravity.

After inputting the setting conditions, the membership function is set, and then the distance and angle between the crops 10 are calculated (steps 104 and 105). That is, the arrangement shown in FIG. 2 has the following regularity common to all the crops 10 that are I × J objects.

Another object always exists in the direction of 90 ° or -90 ° from the end point of the line connecting two adjacent crops 10 between the stocks (rows) as set, and the end point and this object Is equal to the distance between lines (between stocks).

FIG. 3 shows this regularity. Based on this regularity, the following three fuzzy sets are set as antecedents of sequence estimation.

[0021]

(Equation 1) And antecedents are connected by AND to form an estimation rule of three inputs and one output.

FIG. 4 shows membership functions of A, B, and C.

(Equation 2) Is shown.

Here, X1, X2 and θ are antecedent variables. hA (X1) is a triangular type represented by the following equation. If the measured distance is equal to the set value, the grade is set to 1, and if 1/2 or 3/2 of the set value, the grade is set to 0. hA (X1) = 1-│X1 -L1│2 / L1 However, if hA (X1) <0, then hA (X1) = 0. The same applies to hB (X2).

The membership function of C is also triangular, and | 90 | ゜ is grade 1. Then, as shown in FIG. 5, the angle α at which the oblique side of the largest right triangle, which can be defined as a line connecting two adjacent crops 10 as one side, intersects this line is defined as the base point of grade 0.

The membership function is represented by the following equation. hC (θ) = 1− | θ−90 | / (90−α) However, if hC (θ) <0, then hC (θ) = 0.

The reason that α is used as a base point is that among all angles formed between two adjacent ones and any other one, α is an angle other than 90 ° and is the closest angle to 90 °, and a 90 ° lattice arrangement This is because it is easy to make an error in estimating, and it is necessary to make a clear distinction.

The following two types of α can be considered depending on whether the two adjacent pixels are taken in the vertical or horizontal direction of the lattice, and the larger value is adopted.

Α = tan ⁻¹ ((I−1) · L 1 / L 2) when two adjacent ones are taken in the vertical direction, α = tan ⁻¹ (( J-1) · L2 / L1) Which is larger depends on the arrangement setting (L1, L2, I and J).

Next, after calculating the fitness, the crop 1
0 is specified (steps 106 and 107).

That is, for any three of the objects on the binary image, the degree of their closeness to the set arrangement with respect to each other is determined by the degree of conformity ω to the fuzzy set D.
Ask for.

For example, in the objects P, Q, and R that are the crop 10 shown in FIG. 6, first, attention is paid to ∠PQR centered on Q, and the distance between PQ and the set line, the distance between QR and the set line, The degree of conformity between {PQR and | 90 |} is calculated, and their minimum value is defined as the degree of conformity ω1 with the rule as in the following equation.

Ω 1 = h A (X 1) Λh B (X 2) Λh C (θ) where Λ means min. Next, since there is a possibility that the relationship between the distance between the PQ and the QR and the distance between the strips and the stocks may be reversed, the two are interchanged to obtain the fitness ω2, and ω = max {ω1, ω2 、 and ω1 And the maximum value of ω2 is ω.

Similar calculations are performed for ∠QRP and ∠RPQ. In the example of the result shown in FIG.
The case where P is assumed to be between stocks best matches the set arrangement,
The fitness is interpreted to be 0.8. With the above calculation,
For all three combinations, the number of crops is I × J
The crops are specified in order of increasing ω until the number of individual crops is reached.

Here, FIG. 7 shows an example of the result of the present method when cabbage is used. Looking at the original image in FIG. 3A, the growth of the crop 10, which is four vegetables, is not uniform, and the arrangement is slightly disordered. FIG. 2B shows a general RG.
Although this is a binary image obtained by the B image processing method, five objects other than the crop 10 are left as noise.

FIG. 7C shows the result of inputting the barycentric coordinates of the object including the total of nine crops 10 shown in FIG. The center of the circle is the coordinate position, and the solid circle is the object regarded as the crop 10, indicating that four individuals were correctly extracted.

As described above, in the present embodiment, the regularity of the arrangement of the planted crops 10 is used, the ambiguity is flexible, and the arrangement setting and the scale of the arrangement handled in one screen are changed. By using the fuzzy theory that can easily cope with the above, the object having the closest positional relationship to the set crop arrangement is regarded as the crop 10 from the image information in which the crop 10 and the noise (substances other than the crop 10) are mixed. Therefore, only the crop 10 can be extracted.

That is, by adding a method of estimating the crop 10 using the regularity of the arrangement to the image processing method generally used so far, the extraction accuracy of the crop 10 is improved, and fuzzy inference is used. The actual crop 1
Even if the arrangement of 0s is disturbed, the influence is reduced.

Therefore, by employing the method of the present embodiment, it is not necessary to completely remove noise at the stage of image processing, and it is possible to select an algorithm that easily causes the above-described image processing error pattern. Thus, the risk that the crop 10 is erased can be reduced.

In this embodiment, the case where the detection method of the present invention is used to extract the crop 10 has been described. However, the present invention is not limited to this example. Needless to say, it is applicable.

The rules such as the definition of the array rule, the method of setting the membership function, and the method of determining the fitness ω in the fuzzy inference of this embodiment can be changed as appropriate according to the actual state of the array and the purpose of detection. Needless to say.

Furthermore, in the present embodiment, the case where the position of the object which is the crop 10 is represented by the barycentric coordinates and the arrangement of the crop 10 is estimated has been described. May be expressed.

[0042]

As described above, according to the crop detection method of the present invention, the regularity of the arrangement of the planted crops is utilized, the ambiguity is flexible, and the setting change between the strips and the inter-strains is changed. Crops and noise (substances other than crops) by using fuzzy theory that can easily respond to changes in the setting of the number of lines to be processed in one screen and the number of crops per one screen
From the image information where is mixed, the object that has the closest positional relationship to the set crop array is regarded as a crop.
The crop on the screen can be extracted by distinguishing the noise on the screen from the crop.

[Brief description of the drawings]

FIG. 1 is a flowchart for explaining a crop detection method according to the present invention.

FIG. 2 is a diagram schematically showing a setting arrangement in the crop detection method of FIG. 1;

FIG. 3 is a diagram showing the regularity of the arrangement of crops.

FIG. 4 is a diagram showing a membership function.

FIG. 5 is a diagram showing how to determine an angle α.

FIG. 6 is a diagram for explaining fuzzy inference.

FIG. 7 is a diagram showing an example of cabbage extraction by sequence estimation.

[Explanation of symbols]

 10 crops

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Yasuhiro Yokochi 1 Yokagaoka, Toyohira-ku, Sapporo City, Hokkaido Agricultural Experiment Station, Ministry of Agriculture, Forestry and Fisheries (56) References JP-A-3-67377 (JP, A) JP-A-5-165518 (JP, A) JP-A-62-160588 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06T 1/00 G06T 7/00 A01D 43/00-45/00

Claims (1)

(57) [Claims] 1. Use of the regularity of the arrangement of planted crops to change the setting of inter-row and inter-array arrangements and to process within one screen
By using a fuzzy inference method that can easily respond to changes in the number of crops per crop per screen, the positional relationship closest to the set crop arrangement can be obtained from image information in which crops and other objects are mixed. A crop detection method characterized in that only crops are extracted by regarding an object that constitutes a crop as a crop.