A multi-threshold Otsu image segmentation method based on improved chicken swarm optimization algorithm

Abstract

A multi-threshold Otsu image segmentation method based on an improved chicken swarm optimization algorithm is used. The individuals of the chicken swarm algorithm population are initialized by PWLCM chaotic mapping to make them evenly distributed in the search space. The position update of the rooster is improved by introducing a nonlinear weight reduction strategy, so that the entire search space can be traversed better in the early stage and better local convergence can be achieved in the later stage. The hen position update formula is optimized by fractional order G-L and adaptive adjustment of fractional order is introduced to enhance the ability of the algorithm to jump out of the local optimal solution. Considering that the fitness of chicks is poor and the learning space is large, a factor of learning from the global optimal individual is introduced into the position update formula to avoid falling into the local optimum. The present invention can solve the problems of long running time and low precision when Otsu performs multi-threshold segmentation through an improved chicken swarm algorithm, so that it can obtain more accurate thresholds and segmentation efficiency, making the timeliness of foreign body detection more efficient.

Description

Technical Field

[0001] The invention relates to the technical field of image segmentation, and in particular to a multi-threshold Otsu image segmentation method based on an improved chicken swarm optimization algorithm. Background Art

[0002] When using machine vision to detect foreign objects, how to quickly separate foreign objects from the background in the captured image is a key technology, and it has high real-time requirements. Image segmentation, as an important part of computer vision, and the complexity of image segmentation have always been an important research direction for experts and scholars at home and abroad. So far, there is no common segmentation method for all images, but a considerable number of research results and methods have been produced.

[0003] The Otsu segmentation method was proposed by Nobuyukiotsu. It studies an algorithm that uses the maximum value of the inter-class variance of an image to automatically select its optimal threshold. Its main idea is to use the grayscale characteristics of the image to separate the target and background of the image. The larger the inter-class variance, the greater the pixel difference between the background and the target of the image, the more accurate the threshold selection, and the better the efficiency and effect of image segmentation.

[0004] The optimal threshold of the threshold method is set by the pixel value of the image. This means that when using the threshold method, it is necessary to select images with obvious differences between the target and background pixels to achieve a good segmentation effect. For images with rich background pixel distribution, more target pixel areas, and noise interference, the segmentation effect of this method will be significantly reduced. In addition, the operation speed is greatly increased, the efficiency is reduced, and it is not conducive to real-time requirements. Therefore, when facing multi-threshold segmentation, how to improve efficiency while ensuring accuracy is an issue that cannot be ignored and is also a problem worth studying. Summary of the invention

[0005] In response to the above-mentioned technical problems, the present technical solution provides a multi-threshold Otsu image segmentation method based on an improved chicken swarm optimization algorithm. The improved chicken swarm algorithm can solve the problems of long running time and low precision when Otsu performs multi-threshold segmentation, so that it can obtain more accurate thresholds and segmentation efficiency, making the timeliness of foreign body detection more efficient; it can effectively solve the above-mentioned problems.

[0006] The present invention is achieved through the following technical solutions:

[0007] A multi-threshold Otsu image segmentation method based on an improved chicken swarm optimization algorithm reads an image to be threshold segmented, uses the inter-class variance of the multi-threshold Otsu algorithm as a fitness function, and can separate foreign objects and backgrounds in the positioning area through the improved algorithm to achieve the function of foreign object detection; the specific steps include:

[0008] Step 1: Read the image to be threshold segmented, and read the pre-processed image of the wireless charging transmitter positioning area in real time;

[0009] Step 2: Initialize the parameters of the improved chicken swarm algorithm;

[0010] Step 3: Use PWLCM chaotic mapping to initialize the swarm algorithm population particles so that they are evenly distributed in the search space;

[0011] Step 4: Calculate the histogram of the image and set the number of thresholds to be segmented;

[0012] Step 5: Calculate the maximum inter-class variance of the image as the food source of the chickens;

[0013] Step 6: Use the improved formula to update the positions of the rooster, hen and chicks respectively;

[0014] A nonlinear weight reduction strategy is introduced to improve the rooster's position update;

[0015] Adaptive fractional-order GL is introduced to optimize the hen position update formula and introduce adaptive adjustment of fractional order;

[0016] Introduce a factor to learn from the global optimal individual in the chicken position update formula to avoid falling into the local optimum;

[0017] Step 7: Determine whether the current food source is optimal or reaches the number of iterations. If so, proceed to the next step. If not, return to step 5.

[0018] Step 8: Output the maximum threshold;

[0019] Step nine: Perform multi-threshold segmentation on the image.

[0020] Furthermore, the calculation formula of the PWLCM chaotic mapping described in step 3 is:

[0021]

[0022] Wherein, p=0.4, x(1)=rand, and x(t) is the value of the tth iteration.

[0023] Furthermore, the calculation formula of the maximum inter-class variance in step 5 is:

[0024]

[0025] Among them, the threshold combination is [t1, t2, …t K-1 ], divide the image into K categories; is the proportion of each category in the image after segmentation, μ i is the average grayscale of each category, μT is the average grayscale of the image.

[0026] Furthermore, the nonlinear weight reduction strategy described in step 6 is introduced to improve the position update of the rooster. This is because the rooster has the highest fitness value and is the best individual in each group. The introduction of the nonlinear weight reduction strategy makes it have a larger weight in the early stage of iteration, so that it can better traverse the entire search space, and keep a smaller weight in the later stage, so that the population has better local search ability and converges to the global optimal point; the rooster position update formula is:

[0027] x i,j t+1 =λ t *x i,j t *(1+Rand(0,σ 2 )) , (2)

[0028]

[0029]

[0030] Among them, x i,j t+1 represents the position of the rooster at the t+1th iteration, λ t represents the weight of the tth iteration, x i,j t represents the position of the rooster at the tth iteration, Rand(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution; ε represents a very small equilibrium constant to avoid divisor zero; s represents any individual among all roosters except the i-th individual, and the fitness of the i-th rooster is f i , the fitness of the randomly selected rooster s is f s ; t is the current iteration number, t max is the maximum number of iterations, λ max is the initial inertia weight, which is 1.2; min is the inertia weight when the evolution reaches the maximum number of iterations, and its value is 0.1; a and b are adjustment factors, and their value ranges are a=30 and b=0.88.

[0031] Furthermore, the adaptive fractional-order GL described in step 6 takes the first four items of the fractional-order GL to improve the hen position update, and the improved formula is:

[0032]

[0033] The hen position update formula is optimized in fractional order. The original position update formula is: i,j’t+1 =x i,j ’t +a1*Rand*(x r1 ‘t -x i,j ’t )+a2*Rand*(x r2,j ‘t -x i,j ’t )(5);

[0034] Move the item to get:

[0035] x i,j ‘t+1 -x i,j ’t =a1*Rand*(x r1 ‘t -x i,j ’t )+a2*Rand*(x r2,j ‘t -x i,j ’t )(6);

[0036] According to formula (4), taking the first four terms of the fractional order, we can get:

[0037]

[0038] When the fractional order α in equation (7) is 1, equation (6) is combined with equation (7) to obtain the update formula of the fractional hen position:

[0039]

[0040] a2=exp(f r2 -f i )

[0041] Among them, Rand is a random number uniformly distributed between [0,1], and the fitness value of the hen’s partner rooster r1 is f r1 , a1 represents the influence factor of its partner rooster on it, and the fitness value of r2 randomly selected from other roosters and hens is f r2 , a2 is the influence factor of other chickens on it.

[0042] The fractional order α is adaptively adjusted using the position information of the hens. The average distance between hen i and other hens is:

[0043]

[0044] Among them, N is the total number of hen population and D is the spatial dimension.

[0045] Then the evolution factor ω can be expressed as:

[0046]

[0047] d best is the average distance between the global optimal position and other hens, and the maximum average distance of the hen is d max , the minimum value is d min , ω∈[0,1]; when the fractional order α∈[0.5,0.8], the convergence speed of the algorithm is faster; so α can be dynamically adjusted according to the following formula:

[0048]

[0049] Furthermore, in step 6, the factor of learning from the global optimal individual is introduced into the chicken position update formula. The specific improved formula is:

[0050] x i,j '' t+1 =x i,j '' t +E(x m,j '' t -x i,j '' t )+S t (x best,j '' t -x i,j '' t ) (12);

[0051] Among them, the j-th dimension value of the mother hen m position is x m , the influence factor of the mother hen's position on the chick's position is E, which is randomly generated by a random function and its value range is generally (0,2); the update formula of S is t is the current iteration number, t max is the maximum number of iterations, and the value range of S is from 1 to 0;

[0052] Furthermore, the fitness function adopts the inter-class variance of the multi-threshold Otsu algorithm, and the calculation formula is:

[0053]

[0054] Among them, the threshold combination is [t1, t2, …t K-1 ], divide the image into K categories; is the proportion of each category in the image after segmentation, μ i is the average grayscale of each category, μ T is the average grayscale of the image.

[0055] Beneficial Effects

[0056] The multi-threshold Otsu image segmentation method based on the improved chicken swarm optimization algorithm proposed in the present invention has the following beneficial effects compared with the prior art:

[0057] (1) The present invention initializes the individuals of the chicken swarm algorithm population through PWLCM chaotic mapping, so that they are evenly distributed in the search space, thereby improving the global search capability of the population. The position update of the rooster is improved by introducing a nonlinear weight dynamic decrease strategy, and the inertia weight is nonlinearly decreased according to an S-shaped curve; so that the position update of the rooster traverses the entire search space better in the early stage, and has better local convergence in the later stage.

[0058] (2) The present invention optimizes the hen position update formula through fractional-order GL and introduces adaptive adjustment of fractional orders, introduces fractional-order calculus into the hen position update stage of the chicken flock algorithm, and uses its memory to adjust the position of the hen in each iteration. The algorithm can use more historical memory of each hen to search and adaptively adjust the fractional order a according to the position information of the hen; enhance the ability of the algorithm to jump out of the local optimal solution, making the search process more balanced and the search accuracy higher.

[0059] (3) In the chicken group algorithm of the present invention, the fitness value of the chick is the worst. In the original formula, the chick is easily trapped in the local optimum based only on the last position iteration and the influence of the chick's mother hen. Therefore, a factor of learning from the global optimal individual is added to enable the chick to better perform global search and avoid falling into the local optimum. S is a dynamically changing value ranging from 1 to 0. The initial value is large, which can better perform global search, jump out of the local optimum, and move to the vicinity of the global optimum in time. The value becomes smaller in the later stage. Since it is located near the global optimum, it only needs to converge to the local optimum, which can accelerate the convergence speed of the algorithm.

[0060] (4) The improved chicken swarm algorithm of the present invention can solve the problems of long running time and low precision when Otsu performs multi-threshold segmentation, so that it can obtain more accurate thresholds and segmentation efficiency, and can be better applied to the field of foreign object detection in wireless charging. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the segmentation method in Example 1 of the present invention.

[0062] Figure 2 It is a schematic diagram of the iterations in which each algorithm in Embodiment 1 of the present invention converges to the optimal solution.

[0063] Figure 3 Schematic diagram of image segmentation of each algorithm in Example 1 of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Under the premise of not departing from the design concept of the present invention, various modifications and improvements made by ordinary persons in the art to the technical solutions of the present invention should all fall within the protection scope of the present invention.

[0065] Embodiment 1:

[0066] like Figure 1 As shown in the figure, a multi-threshold Otsu image segmentation method based on an improved chicken swarm optimization algorithm is used to read the image to be threshold segmented, and the inter-class variance of the multi-threshold Otsu algorithm is used as the fitness function. The foreign matter and the background in the positioning area can be separated by the improved algorithm to realize the function of foreign matter detection; the specific steps include:

[0067] Step 1: Read the image to be thresholded and read the pre-processed image of the wireless charging transmitter positioning area in real time.

[0068] Step 2: Initialize the parameters of the improved chicken swarm algorithm.

[0069] Step 3: Use PWLCM chaotic mapping to initialize the population particles of the chicken swarm algorithm so that they are evenly distributed in the search space; the calculation formula of PWLCM chaotic mapping is:

[0070]

[0071] Wherein, p=0.4, x(1)=rand, and x(t) is the value of the tth iteration.

[0072] Step 4: Calculate the histogram of the image and set the number of thresholds to be segmented.

[0073] Step 5: Calculate the maximum inter-class variance of the image as the food source of the chickens; the calculation formula for the maximum inter-class variance is:

[0074]

[0075] Among them, the threshold combination is [t1, t2, …t K-1 ], divide the image into K categories; is the proportion of each category in the image after segmentation, μ i is the average grayscale of each category, μ T is the average grayscale of the image.

[0076] Step 6: Use the improved formula to update the positions of the rooster, hen and chicks respectively;

[0077] The nonlinear weight reduction strategy is introduced to improve the position update of the rooster. The rooster has the highest fitness value and is the best individual in each group. The nonlinear weight reduction strategy is introduced to make it have a larger weight in the early stage of iteration, so that it can better traverse the entire search space, and keep a smaller weight in the later stage, so that the population has better local search ability and converges to the global optimal point. The position update formula of the rooster is:

[0078] x i,j t+1 =λ t *x i,j t *(1+Rand(0,σ 2 ))(2);

[0079]

[0080] Among them, x i,j t+1 represents the position of the rooster at the t+1th iteration, λ t represents the weight of the tth iteration, x i,j t represents the position of the rooster at the tth iteration, Rand(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution; ε represents a very small equilibrium constant to avoid divisor zero; s represents any individual among all roosters except the i-th individual, and the fitness of the i-th rooster is f i , the fitness of the randomly selected rooster s is f s ; t is the current iteration number, t max is the maximum number of iterations, λ max is the initial inertia weight, which is 1.2; min is the inertia weight when the evolution reaches the maximum number of iterations, and its value is 0.1; a and b are adjustment factors, and their value ranges are a=30 and b=0.88.

[0081] Adaptive fractional-order GL is introduced to optimize the hen position update formula and introduce adaptive adjustment of fractional order; the first four terms of fractional-order GL are taken to improve the hen position update, and the improved formula is:

[0082]

[0083] The hen position update formula is optimized in fractional order. The original position update formula is: i,j ’t+1 =x i,j ’t +a1*Rand*(x r1 ‘t-x i,j ’t )+a2*Rand*(x r2,j ‘t -x i,j ’t )(5);

[0084] Move the item to get:

[0085] x i,j ‘t+1 -x i,j ’t =a1*Rand*(x r1 ‘t -x i,j ’t )+a2*Rand*(x r2,j ‘t -x i,j ’t )(6);

[0086] According to formula (4), taking the first four terms of the fractional order, we can get:

[0087]

[0088] When the fractional order α in equation (7) is 1, equation (6) is combined with equation (7) to obtain the update formula of the fractional hen position:

[0089]

[0090] a2=exp(f r2 -f i )

[0091] Among them, Rand is a random number uniformly distributed between [0,1], and the fitness value of the hen’s partner rooster r1 is f r1 , a1 represents the influence factor of its partner rooster on it, and the fitness value of r2 randomly selected from other roosters and hens is f r2 , a2 is the influence factor of other chickens on it.

[0092] The fractional order α is adaptively adjusted using the position information of the hens. The average distance between hen i and other hens is:

[0093]

[0094] Among them, N is the total number of hen population and D is the spatial dimension.

[0095] Then the evolution factor ω can be expressed as:

[0096]

[0097] d best is the average distance between the global optimal position and other hens, and the maximum average distance of the hen is d max , the minimum value is d min , ω∈[0,1]; when the fractional order α∈[0.5,0.8], the algorithm converges faster, so α can be dynamically adjusted according to the following formula:

[0098]

[0099] Introduce the factor of learning from the global optimal individual in the chicken's position update formula to avoid falling into the local optimum; Introduce the factor of learning from the global optimal individual in the chicken's position update formula. The specific improved formula is:

[0100] x i,j '' t+1 =x i,j '' t +E(x m,j '' t -x i,j '' t )+S t (x best,j '' t -x i,j '' t ) (12);

[0101] Among them, the j-th dimension value of the mother hen m position is x m , the influence factor of the mother hen's position on the chick's position is E, which is randomly generated by a random function and its value range is generally (0,2); the update formula of S is t is the current iteration number, t max is the maximum number of iterations, and the value of S ranges from 1 to 0.

[0102] Step 7: Determine whether the current food source is optimal or reaches the number of iterations. If so, proceed to the next step. If not, return to step 5.

[0103] Step 8: Output the maximum threshold;

[0104] Step nine: Perform multi-threshold segmentation on the image.

[0105] The fitness function of this embodiment adopts the inter-class variance of the multi-threshold Otsu algorithm, and the calculation formula is:

[0106]

[0107] Among them, the threshold combination is [t1, t2, …t K-1 ], divide the image into K categories; is the proportion of each category in the image after segmentation, μ i is the average grayscale of each category, μ T is the average grayscale of the image.

[0108] The inventor selected human images as a test, evaluated the algorithm by fitness value, and compared it with FA-Otsu and Pso-Otsu algorithms respectively. Figure 3 (a) is the original image after grayscale processing, (b) is the image segmented by FA-Otsu, (c) is the image segmented by PSO-Otsu, and (d) is the image segmented by the algorithm in this paper. Figure 2 It can be seen that this algorithm converges to the optimal point after only 4 iterations, while FA-Otsu converges to the local optimum after about 30 iterations, and finally converges to the global optimal solution after about 45 iterations, and PSO-Otsu converges to the optimal solution after about 12 iterations. Compared with the other two algorithms, this algorithm has greatly improved the efficiency of segmentation and can solve the problems of long running time and low accuracy when Otsu performs multi-threshold segmentation.

[0109] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes, substitutions and improvements within the technical scope disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A multi-threshold Otsu image segmentation method based on an improved chicken swarm optimization algorithm, characterized in that: Read the image to be thresholded and use the inter-class variance of the multi-threshold Otsu algorithm as the fitness function. The improved algorithm can separate the foreign objects and the background in the positioning area, thus realizing the function of foreign object detection. The specific steps include: Step 1: Read the image to be threshold segmented, and read the pre-processed image of the wireless charging transmitter positioning area in real time; Step 2: Initialize the parameters of the improved chicken swarm algorithm; Step 3: Use PWLCM chaotic mapping to initialize the swarm algorithm population particles so that they are evenly distributed in the search space; Step 4: Calculate the histogram of the image and set the number of thresholds to be segmented; Step 5: Calculate the maximum inter-class variance of the image as the food source of the chickens; Step 6: Use the improved formula to update the positions of the rooster, hen and chicks respectively; Step 6.1: Introduce a nonlinear weight reduction strategy to improve the position update of the rooster; Since the rooster has the highest fitness value and is the best individual in each group, a nonlinear weight reduction strategy is introduced to give it a larger weight in the early stage of iteration to better traverse the entire search space, while keeping a smaller weight in the later stage so that the population has better local search ability and converges to the global optimal point; the rooster position update formula is: x i,j t+1 =λ t *x i,j t *(1+Rand(0,σ 2 )),(2) Among them, x i,j t+1 represents the position of the rooster at the t+1th iteration, λ t represents the weight of the tth iteration, x i,j t represents the position of the rooster at the tth iteration, Rand(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution; ε represents a very small equilibrium constant to avoid divisor zero; s represents any individual among all roosters except the i-th individual, and the fitness of the i-th rooster is f i , the fitness of the randomly selected rooster s is f s ; t is the current iteration number, t max is the maximum number of iterations, λ max is the initial inertia weight, which is 1.2; min is the inertia weight when the evolution reaches the maximum number of iterations, and its value is 0.1; a and b are adjustment factors, and their value ranges are a=30, b=0.88; Step 6.2: Introduce adaptive fractional-order GL to optimize the hen position update formula and introduce adaptive adjustment of fractional order; The adaptive fractional-order GL takes the first four items of the fractional-order GL to improve the hen position update, and the improved formula is: The hen position update formula is optimized in fractional order. The original position update formula is: i,j ’t+1 =x i,j ’t +a1*Rand*(x r1 ‘t -x i,j ’t )+a2*Rand*(x r2,j ‘t -x i,j ’t )(5); Move the item to get: x i,j ‘t+1 -x i,j ’t =a1*Rand*(x r1 ‘t -x i,j ’t )+a2*Rand*(x r2,j ‘t -x i,j ’t )(6); According to formula (4), taking the first four terms of the fractional order, we can get: When the fractional order α in equation (7) is 1, equation (6) is combined with equation (7) to obtain the update formula of the fractional hen position: a2=exp(f r2 -f i ) Among them, Rand is a random number uniformly distributed between [0,1], and the fitness value of the hen’s partner rooster r1 is f r1 , a1 represents the influence factor of its partner rooster on it, and the fitness value of r2 randomly selected from other roosters and hens is f r2 , a2 is the influence factor of other chickens on it; The fractional order α is adaptively adjusted using the position information of the hens. The average distance between hen i and other hens is: Where N is the total number of hen populations and D is the spatial dimension; the evolution factor ω can be expressed as: d best is the average distance between the global optimal position and other hens, and the maximum average distance of the hen is d max , the minimum value is d min , ω∈[0,1]; when the fractional order α∈[0.5,0.8], the algorithm converges faster, so α can be dynamically adjusted according to the following formula: Step 6.3: Introduce the factor of learning from the global optimal individual in the chicken's position update formula to avoid falling into the local optimum; Introduce the factor of learning from the global optimal individual in the chicken's position update formula. The specific improved formula is: x i,j ‘’t+1 =x i,j ‘’t +E(x m,j ‘’t -x i,j ’‘t )+S t (x best,j ‘’t -x i,j ‘’t )(12); Among them, the j-th dimension value of the mother hen m position is x m , the influence factor of the mother hen's position on the chick's position is E, which is randomly generated by a random function and its value range is generally (0,2); the update formula of S is t is the current iteration number, t max is the maximum number of iterations, and the value range of S is from 1 to 0; x best,j ‘’t The optimal position in the chick group; Step 7: Determine whether the current food source is optimal or reaches the number of iterations. If so, proceed to the next step. If not, return to step 5. Step 8: Output the maximum threshold; Step nine: Perform multi-threshold segmentation on the image.

2. The multi-threshold Otsu image segmentation method based on the improved chicken swarm optimization algorithm according to claim 1 is characterized in that: The calculation formula of the PWLCM chaotic mapping described in step 3 is: Wherein, p=0.4, x(1)=rand, and x(t) is the value of the tth iteration.

3. The multi-threshold Otsu image segmentation method based on the improved chicken swarm optimization algorithm according to claim 1 is characterized in that: The calculation formula for the maximum inter-class variance in step 5 is: Among them, the threshold combination is [t1, t2, …t K-1 ], divide the image into K categories; is the proportion of each category in the image after segmentation, μ i is the average grayscale of each category, μ T is the average grayscale of the image.

4. The multi-threshold Otsu image segmentation method based on the improved chicken swarm optimization algorithm according to claim 1 is characterized in that: The fitness function uses the inter-class variance of the multi-threshold Otsu algorithm, and the calculation formula is: Among them, the threshold combination is [t1, t2, …t K-1 ], divide the image into K categories; is the proportion of each category in the image after segmentation, μ i is the average grayscale of each category, μ T is the average grayscale of the image.

Images