Citrus leaf disease segmentation method based on image processing

Abstract

The invention provides a citrus leaf disease spot image segmentation method based on an image processing citrus leaf disease segmentation method, belongs to the technical field of agricultural plant protection, and comprises an image segmentation algorithm based on Otsu threshold segmentation, a Cb component graph segmentation method based on a YCrCbCr color space, and a multi-threshold segmentation method based on GWO-PSO and a disease spot segmentation algorithm. Segmenting the Cb component graph into a background image and a target image by adopting an Otsu threshold value, and performing morphological processing on the segmented target image to obtain a citrus leaf region in the image; in order to solve the problem that the color of the lesion area on the citrus leaf is inconsistent, a gray wolf-particle swarm dual-threshold adaptive algorithm is adopted to segment the healthy area and the lesion area of the leaf, and the lesion degree of the citrus tree can be roughly evaluated on the basis. The citrus leaf disease segmentation method based on image processing is fast in algorithm convergence rate, high in precision and strong in generalization ability, can be applied to multi-threshold segmentation of leaf disease spots, and provides powerful information support for subsequent citrus tree lesion identification and automatic pesticide application.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural plant protection, and more specifically, particularly relates to a citrus leaf disease segmentation method based on image processing. Background Art

[0002] Crop lesion area segmentation is a key issue in plant disease severity assessment and disease type identification. Common lesion image segmentation methods include threshold segmentation, region segmentation, and watershed segmentation. The threshold segmentation method has the advantages of simple and fast operation, and has been widely studied and applied in segmentation. In order to take into account the efficiency and accuracy of the segmentation algorithm, the present invention provides a citrus leaf disease segmentation method based on image processing, which can accurately segment the anthracnose disease area of ​​citrus leaves. It is superior to traditional algorithms in terms of accuracy, processing speed and robustness, and can obtain better disease segmentation effects. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present invention provides a citrus leaf disease segmentation method based on image processing, which obtains the citrus leaf area in the image through a secondary segmentation method of first coarse segmentation and then fine segmentation, so as to solve the problem of pesticide waste caused by the difficulty in determining the diseased area in the prior art.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a citrus leaf disease segmentation method based on image processing is used to segment the healthy area and the diseased area of ​​citrus leaves, based on which the degree of citrus tree disease can be evaluated. The segmentation method specifically includes the following steps:

[0005] S1: Acquire citrus anthracnose images in an orchard using a visible light camera and crop the image area containing the diseased leaves.

[0006] S2: Select a flat leaf image, remove part of the background using screenshot software, and use median filtering and edge sharpening methods for image preprocessing. Finally, only one leaf containing citrus anthracnose is retained in each image to improve the subsequent segmentation efficiency.

[0007] S3: Use Otsu threshold to perform rough segmentation on the image, the goal is to segment the target area (leaves) and the background area (excluding leaves).

[0008] S4: The multi-threshold segmentation method based on GWO-PSO is used to finely segment the target area and segment the diseased area in the leaf.

[0009] S5: According to the disease segmentation results, the number of pixels in the segmented lesion area is divided by the number of pixels in the target area as the lesion assessment parameter to calculate the citrus leaf lesion rate. The calculation method is:

[0010]

[0011] Where DD is the current leaf disease diagnosis result; S d is the total number of pixels of the lesion; S h is the number of pixels in the healthy area of ​​the leaf.

[0012] As a further improvement of the present invention, in step S3, the Otsu threshold segmentation method includes the following steps:

[0013] S31: Convert the original RGB model disease image to the YCbCr color model. The calculation method is:

[0014]

[0015] Where R is the R component of the citrus leaf disease image; G is the G component of the citrus leaf disease image; B is the B component of the citrus leaf disease image; Y represents the brightness and concentration of the citrus leaf disease image color, also called grayscale; Cb represents the blue concentration offset of the citrus leaf disease image color; Cr represents the red concentration offset of the citrus leaf disease image color.

[0016] S32: extracting the Y, Cb, and Cr component histograms of the defect image, and analyzing whether the component histograms show a bimodal feature.

[0017] S33: According to the histogram calculation results, the citrus anthracnose diseased leaf image has an obvious bimodal feature in the Cb component, and thus the Cb component grayscale image of the diseased image is selected for Otsu threshold segmentation.

[0018] S34: Since the diseased part inside the citrus leaf area after Otsu threshold segmentation is easily segmented as the background area, the morphological closing operation is used to eliminate the part of the leaf that is misjudged as the background area to obtain the target leaf area including the diseased part. The calculation method is:

[0019]

[0020] In the formula, B is the structural element; A is the disease image to be processed; Indicates that the structural element B is used to dilate the image A; Θ is the erosion operation.

[0021] S35: Using the diseased image after Otsu threshold segmentation and closing operation as a basis, filtering out the background part in the original image, and completing the segmentation of the leaf area and the background area of ​​the image.

[0022] As a further improvement of the present invention, in step S4, the multi-threshold segmentation method based on GWO-PSO includes the following steps:

[0023] S41: Particle swarm optimization (PSO) algorithm is used to find the optimal solution through group collaboration and competition. The speed v of the dth iteration of the i-th particle is id The update calculation is:

[0024] v id =w×v id-1 +c 1 × 1 ×(p id -x id )+c 2 × 2 ×(p gd -x id )

[0025] Where w is the inertia coefficient; v id-1 is the velocity of the last (d-1th) iteration of the ith particle; c 1 is the individual learning factor; c 2 is the group learning factor; p id is the extreme value of the individual's d-th iteration; x id is the position of the dth iteration; p gd is the extreme value of the d-th iteration of the group; r 1 、r 2 is a random number between [0,1].

[0026] Specifically, the position update calculation method of PSO is:

[0027] x id+1 =x id +v id

[0028] In the formula, x id+1 is the position of the next (d+1) iteration.

[0029] S42: Using the Grey Wolf Optimization (GWO) algorithm, we design three types of ruling class wolves and ordinary wolves. The ruling class wolves are divided into α wolves, β wolves, δ wolves, and ordinary wolves that make up the wolf pack, which correspond to the highest fitness, the second highest, the third highest, and other solutions respectively. The specific calculation steps of the GWO algorithm are:

[0030] (1) Calculate the distance D between the individual and the prey using the following calculation method:

[0031]

[0032] In the formula, is the synergy coefficient vector; is the position vector of the prey; is the position vector of the gray wolf.

[0033] Specifically, the synergy coefficient vector The calculation method is:

[0034]

[0035] In the formula, Get a random number between [0,1] modulo.

[0036] Specifically, the distance calculation formula between the gray wolf individual and the α wolf, β wolf, and δ wolf is:

[0037]

[0038] In the formula, is the distance between the wolf and the prey; is the distance between β wolf and prey; is the distance between the wolf and the prey; is the synergy coefficient vector; is the position vector of the α wolf prey; is the position vector of β wolf prey; is the position vector of the wolf’s prey.

[0039] (2) Calculate the gray wolf's position update:

[0040]

[0041] In the formula, is the next position vector of the gray wolf; is the synergy coefficient vector.

[0042] Specifically, the synergy coefficient vector The calculation method is:

[0043]

[0044] In the formula, Get a random number between [0,1] modulo.

[0045] Specifically, the step length and direction of an ordinary wolf in a wolf pack moving toward wolf α, wolf β, and wolf δ are calculated as follows:

[0046]

[0047] In the formula, The direction in which the normal wolf moves towards the alpha wolf; The direction in which the normal wolf moves towards the beta wolf; The direction in which the common wolf moves toward the delta wolf; is the synergy coefficient vector.

[0048] (3) Calculate the next position of the common wolf The calculation method is:

[0049]

[0050] Specifically, the coefficient As an adjustable parameter, the gray wolf algorithm is optimized by allocating different proportion coefficients of α wolf, β wolf and δ wolf. The new position allocation algorithm allocates the proportions of α wolf, β wolf and δ wolf at a ratio of 50%, 30% and 20%, and the calculation method is:

[0051]

[0052] S43: Use the Grey Wolf Optimization Algorithm (GWO) and the Particle Swarm Optimization Algorithm (PSO) to update the position. Specifically, the Grey Wolf Algorithm is used as the main body, and the position update is learned from the global optimal position and the individual historical optimal position at the same time. The position update calculation method is:

[0053]

[0054] Where w is the inertia coefficient; is the individual historical optimal; c 1 is the individual learning factor; c 2 is the group learning factor; is the current location; The previous step position.

[0055] Specifically, the value range of w is [0.4, 0.9], and the calculation method of w is:

[0056]

[0057] In the formula, w max is the maximum value; w min is the minimum value; i represents the i-th value; i max The maximum number of values.

[0058] S44: Kapur entropy is used for multi-threshold segmentation based on GWO-PSO to obtain good stability and segmentation effect.

[0059] The specific calculation steps of multi-threshold segmentation are:

[0060] (1) Calculate the grayscale probability distribution p of the disease image i , calculated as:

[0061]

[0062] Where n i is the number of pixels with gray value i; i is the pixel with gray value i; N is the total number of pixels; L is the maximum gray level.

[0063] (2) For the multi-level thresholding problem, the j-th Kapur entropy H j can be described as:

[0064]

[0065] Where ti is the ith threshold; t j+1 is the j+1th threshold; p j is the grayscale probability of the number of pixels with grayscale value j; w j is the probability distribution of the j-th class image.

[0066] Specifically, w j The calculation method is:

[0067]

[0068] (3) The optimal m-1 level threshold is obtained by maximizing the objective function. The calculation method is:

[0069]

[0070] In the formula, H i is the i-th threshold; m is the number of thresholds.

[0071] S44: Design the algorithm end condition of the multi-threshold segmentation method based on GWO-PSO. Specifically, if any of the following conditions is met, the iteration is exited and the optimal solution is output:

[0072] 1. The number of iterations is equal to the maximum number of iterations;

[0073] 2. The difference between the maximum applicable function values ​​for 6 consecutive times is less than 10 -6 .

[0074] The present invention provides a citrus leaf disease segmentation method based on image processing. Compared with the existing fruit tree leaf disease observation method, the present invention has the following beneficial effects:

[0075] 1. The present invention adopts the cascade algorithm idea of ​​secondary segmentation. In the second segmentation, in order to improve the accuracy of lesion area segmentation, a multi-threshold segmentation algorithm is adopted. However, when the threshold segmentation method is extended to multi-level thresholds, its computational complexity explodes. In order to reduce the computational complexity, and in view of the good optimization performance and fast convergence speed of the gray wolf-particle swarm optimization algorithm, the present invention introduces the gray wolf-particle swarm optimization algorithm in the secondary segmentation to obtain the image multi-segmentation threshold.

[0076] 2. The present invention can accurately segment the areas with inconsistent colors of the diseased areas on citrus leaves to improve the utilization rate of pesticides in the subsequent application process. The accuracy of leaf disease segmentation is improved by the primary segmentation algorithm based on Otsu and the secondary multi-threshold image segmentation algorithm based on GWO-PSO. The leaf disease segmentation algorithm based on GWO-PSO can effectively overcome the problem of the PSO algorithm falling into the local optimal solution, making the disease segmentation method more robust. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A flowchart of the steps of a citrus leaf disease segmentation method based on image processing provided by the present invention;

[0078] Figure 2 The GWO-PSO algorithm flow chart provided by the present invention;

[0079] Figure 3 A schematic diagram of the cascade image segmentation process provided by the present invention;

[0080] Figure 4 This is a diagram of the segmentation effect of citrus leaf diseases provided by the present invention. DETAILED DESCRIPTION

[0081] The following is a further detailed description of the embodiments of the present invention in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0082] In the description of the present invention, unless otherwise specified, "plurality" means two or more than two; the orientations or positional relationships indicated by the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0083] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the words "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0084] Combine the following Figure 1-Figure 4The present invention describes a disease segmentation method. A citrus leaf disease segmentation method based on image processing is used to segment the healthy area and the diseased area of ​​the citrus leaf, based on which the degree of citrus tree disease can be evaluated. The segmentation method specifically includes the following steps:

[0085] S1: Acquire citrus anthracnose images in an orchard using a visible light camera and crop the image area containing the diseased leaves.

[0086] S2: Select a flat leaf image, remove part of the background using screenshot software, and use median filtering and edge sharpening methods for image preprocessing. Finally, only one leaf containing citrus anthracnose is retained in each image to improve the subsequent segmentation efficiency.

[0087] S3: Use Otsu threshold to perform rough segmentation on the image, the goal is to segment the target area (leaves) and the background area (excluding leaves).

[0088] Further, as a further improvement of the present invention, in step S3, the Otsu threshold segmentation method includes the following steps:

[0089] S31: Convert the original RGB model disease image to the YCbCr color model. The calculation method is:

[0090]

[0091] Where R is the R component of the citrus leaf disease image; G is the G component of the citrus leaf disease image; B is the B component of the citrus leaf disease image; Y represents the brightness and concentration of the citrus leaf disease image color, also called grayscale; Cb represents the blue concentration offset of the citrus leaf disease image color; Cr represents the red concentration offset of the citrus leaf disease image color.

[0092] S32: extracting the Y, Cb, and Cr component histograms of the defect image, and analyzing whether the component histograms show a bimodal feature.

[0093] S33: According to the histogram calculation results, the citrus anthracnose diseased leaf image has an obvious bimodal feature in the Cb component, and thus the Cb component grayscale image of the diseased image is selected for Otsu threshold segmentation.

[0094] S34: Since the diseased part inside the citrus leaf area after Otsu threshold segmentation is easily segmented as the background area, the morphological closing operation is used to eliminate the part of the leaf that is misjudged as the background area to obtain the target leaf area including the diseased part. The calculation method is:

[0095]

[0096] In the formula, B is the structural element; A is the disease image to be processed; Indicates that the structural element B is used to dilate the image A; Θ is the erosion operation.

[0097] S35: Using the diseased image after Otsu threshold segmentation and closing operation as a basis, filtering out the background part in the original image, and completing the segmentation of the leaf area and the background area of ​​the image.

[0098] S4: The multi-threshold segmentation method based on GWO-PSO is used to finely segment the target area and segment the diseased area in the leaf.

[0099] Further, as a further improvement of the present invention, in step S4, the multi-threshold segmentation method based on GWO-PSO includes the following steps:

[0100] S41: Particle swarm optimization (PSO) algorithm is used to find the optimal solution through group collaboration and competition. The speed v of the dth iteration of the i-th particle is id The update calculation is:

[0101] v id =w×v id-1 +c 1 × 1 ×(p id -x id )+c 2 × 2 ×(p gd -x id )

[0102] Where w is the inertia coefficient; v id-1 is the velocity of the last (d-1th) iteration of the ith particle; c 1 is the individual learning factor; c 2 is the group learning factor; p id is the extreme value of the individual's d-th iteration; x id is the position of the dth iteration; p gd is the extreme value of the d-th iteration of the group; r 1 、r 2 is a random number between [0,1].

[0103] Specifically, the position update calculation method of PSO is:

[0104] x id+1 =x id +v id

[0105] In the formula, x id+1 is the position of the next (d+1) iteration.

[0106] S42: Using the Grey Wolf Optimization (GWO) algorithm, we design three types of ruling class wolves and ordinary wolves. The ruling class wolves are divided into α wolves, β wolves, δ wolves, and ordinary wolves that make up the wolf pack, which correspond to the highest fitness, the second highest, the third highest, and other solutions respectively. The specific calculation steps of the GWO algorithm are:

[0107] (1) Calculate the distance between the individual and the prey The calculation method is:

[0108]

[0109] In the formula, is the synergy coefficient vector; is the position vector of the prey; is the position vector of the gray wolf.

[0110] Specifically, the synergy coefficient vector The calculation method is:

[0111]

[0112] In the formula, Get a random number between [0,1] modulo.

[0113] Specifically, the distance calculation formula between the gray wolf individual and the α wolf, β wolf, and δ wolf is:

[0114]

[0115] In the formula, is the distance between the wolf and the prey; is the distance between β wolf and prey; is the distance between the wolf and the prey; is the synergy coefficient vector; is the position vector of the α wolf prey; is the position vector of β wolf prey; is the position vector of the wolf’s prey.

[0116] (2) Calculate the gray wolf's position update:

[0117]

[0118] In the formula, is the next position vector of the gray wolf; is the synergy coefficient vector.

[0119] Specifically, the synergy coefficient vector The calculation method is:

[0120]

[0121] In the formula, Get a random number between [0,1] modulo.

[0122] Specifically, the step length and direction of an ordinary wolf in a wolf pack moving toward wolf α, wolf β, and wolf δ are calculated as follows:

[0123]

[0124] In the formula, The direction in which the normal wolf moves towards the alpha wolf; The direction in which the normal wolf moves towards the beta wolf; The direction in which the common wolf moves toward the delta wolf; is the synergy coefficient vector.

[0125] (3) Calculate the next position of the common wolf The calculation method is:

[0126]

[0127] Specifically, the coefficient As an adjustable parameter, the gray wolf algorithm is optimized by allocating different proportion coefficients of α wolf, β wolf and δ wolf. The new position allocation algorithm allocates the proportions of α wolf, β wolf and δ wolf at a ratio of 50%, 30% and 20%, and the calculation method is:

[0128]

[0129] S43: Use the Grey Wolf Optimization Algorithm (GWO) and the Particle Swarm Optimization Algorithm (PSO) to update the position. Specifically, the Grey Wolf Algorithm is used as the main body, and the position update is learned from the global optimal position and the individual historical optimal position at the same time. The position update calculation method is:

[0130]

[0131] Where w is the inertia coefficient; is the individual historical optimal; c 1 is the individual learning factor; c 2 is the group learning factor; is the current location; The previous step position.

[0132] Specifically, the value range of w is [0.4, 0.9], and the calculation method of w is:

[0133]

[0134] In the formula, w max is the maximum value; w minis the minimum value; i represents the i-th value; i max The maximum number of values.

[0135] S44: Kapur entropy is used for multi-threshold segmentation based on GWO-PSO to obtain good stability and segmentation effect.

[0136] The specific calculation steps of multi-threshold segmentation are:

[0137] (1) Calculate the grayscale probability distribution p of the disease image i , calculated as:

[0138]

[0139] Where n i is the number of pixels with gray value i; i is the pixel with gray value i; N is the total number of pixels; L is the maximum gray level.

[0140] (2) For the multi-level thresholding problem, the j-th Kapur entropy H j can be described as:

[0141]

[0142] Where ti is the ith threshold; t j+1 is the j+1th threshold; p j is the grayscale probability of the number of pixels with grayscale value j; w j is the probability distribution of the j-th class image.

[0143] Specifically, w j The calculation method is:

[0144]

[0145] (3) The optimal m-1 level threshold is obtained by maximizing the objective function. The calculation method is:

[0146]

[0147] In the formula, H i is the i-th threshold; m is the number of thresholds.

[0148] S44: Design the algorithm end condition of the multi-threshold segmentation method based on GWO-PSO. Specifically, if any of the following conditions is met, the iteration is exited and the optimal solution is output:

[0149] 1. The number of iterations is equal to the maximum number of iterations;

[0150] 2. The difference between the maximum applicable function values ​​for 6 consecutive times is less than 10 -6 .

[0151] S5: According to the disease segmentation results, the number of pixels in the segmented lesion area is divided by the number of pixels in the target area as the lesion assessment parameter to calculate the citrus leaf lesion rate. The calculation method is:

[0152]

[0153] Where DD is the current leaf disease diagnosis result; S d is the total number of pixels of the lesion; S h is the number of pixels in the healthy area of ​​the leaf.

[0154] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A citrus leaf disease segmentation method based on image processing, characterized in that: It includes an image segmentation algorithm based on Otsu threshold segmentation, a Cb component map segmentation method based on YCrCbCr color space, and a lesion segmentation algorithm based on a multi-threshold segmentation method of GWO-PSO.

2. The citrus leaf disease image segmentation method based on image processing according to claim 1 is characterized in that: The specific steps are: S1: Obtain an image of citrus anthracnose in an orchard using a visible light camera and crop the image area containing the diseased leaves; S2: Select the flattened image of leaves, remove part of the background through screenshot software, and use median filtering and edge sharpening methods for image preprocessing. Finally, only one leaf containing citrus anthracnose is retained in each image to improve the subsequent segmentation efficiency; S3: Use Otsu threshold to roughly segment the image, the goal is to segment the target area (leaves) and the background area (excluding leaves); S4: The target area is finely segmented using the multi-threshold segmentation method based on GWO-PSO to segment the diseased area in the leaf; S5: According to the disease segmentation results, the number of pixels in the segmented lesion area divided by the number of pixels in the target area is used as the lesion assessment parameter to calculate the citrus leaf lesion rate.

3. The citrus leaf spot image segmentation method according to claim 1, characterized in that: The original RGB model disease image is converted to the YCbCr color model, and the calculation method is: Where R is the R component of the citrus leaf disease image; G is the G component of the citrus leaf disease image; B is the B component of the citrus leaf disease image; Y represents the brightness and concentration of the citrus leaf disease image color, also called grayscale; Cb represents the blue concentration offset of the citrus leaf disease image color; Cr represents the red concentration offset of the citrus leaf disease image color.

4. The citrus leaf spot image segmentation method according to claim 1, characterized in that: The morphological closing operation is used to obtain the target area of ​​the leaf including the diseased part. The calculation method is: In the formula, B is the structural element; A is the disease image to be processed; Indicates that the structural element B is used to dilate the image A; Θ is the erosion operation.

5. The citrus leaf spot image segmentation method according to claim 1, characterized in that: The particle swarm optimization algorithm (PSO) is used to calculate the velocity v of the dth iteration of the i-th particle. id The update method is: v id =w×v id-1 +c1×r1×(p id -x id )+c2×r2×(p gd -x id ) Where w is the inertia coefficient; v id-1 is the speed of the last iteration (d-1th iteration) of the ith particle; c1 is the individual learning factor; c2 is the group learning factor; p id is the extreme value of the individual's d-th iteration; x id is the position of the dth iteration; p gd is the extreme value of the d-th iteration of the population; r1 and r2 are random numbers between [0,1].

6. The citrus leaf disease image segmentation method based on image processing according to claim 1, characterized in that: The Grey Wolf Optimization Algorithm (GWO) is combined with the Particle Swarm Optimization Algorithm (PSO) to update the position. The position update calculation method is: Where w is the inertia coefficient; is the individual historical optimum; c1 is the individual learning factor; c2 is the group learning factor; is the current location; The previous step position.

7. The citrus leaf disease image segmentation method based on image processing according to claim 1, characterized in that: According to the disease segmentation results, the number of pixels in the segmented lesion area divided by the number of pixels in the target area is used as the lesion assessment parameter to calculate the citrus leaf lesion rate. The calculation method is: Where DD is the current leaf disease diagnosis result; S d is the total number of pixels of the lesion; S h is the number of pixels in the healthy area of ​​the leaf.

Images