CN105354585A - Improved cat swarm algorithm based target extraction and classification method - Google Patents

Abstract

The invention provides a method for object extraction and classification based on the improved cat group algorithm. The traditional swarm intelligence algorithm is too complex in image target extraction and classification, and it is easy to fall into local optimum, resulting in "premature". For example, the cat swarm algorithm is a typical swarm intelligence algorithm, but it has the disadvantages of long running time and low accuracy in the image processing of big data. For this reason, the present invention proposes an improved cat swarm algorithm for the shortcomings of the cat swarm algorithm. We have increased the inertia weight coefficient and the acceleration coefficient in the tracking mode of the algorithm, which has improved the running speed of the algorithm and shortened the running time. And the improved cat group algorithm is applied to the extraction and classification of the target object, that is: first input the image, preprocess the image, and threshold it into a binary image, extract the target image of interest, and calculate the four-point value of the target object. features, form a new feature vector, and finally use the improved cat group algorithm to classify. This method can not only improve the calculation speed, but also improve the accuracy of target object extraction and classification.

Description

A Method of Target Extraction and Classification Based on Improved Cat Swarm Algorithm

technical field

The invention relates to swarm intelligence, bionic computing and pattern recognition technology, in particular to a method for object extraction and classification based on the improved cat group algorithm. This method has broad application prospects in image recognition, pattern classification, object tracking and other fields.

Background technique

The hotspots and difficulties of target object extraction and classification based on clustering methods in the fields of image processing and pattern recognition. Therefore, many scholars devote themselves to researching this hot spot and difficulty, and propose a series of clustering algorithms at the same time. For example, the well-known k-means algorithm (Terada Yoshikazu. Strong Consistency of Reduced K-means Clustering [J]. Scandinavian journal of statistics, 41 (4), 2014) and other traditional clustering algorithms. However, this algorithm has a big disadvantage, that is, it is sensitive to the selection of cluster centers, and it is easy to fall into local optimum prematurely. In order to overcome the shortcomings of traditional clustering algorithms, many swarm intelligence algorithms that simulate animal behavior have also been used to study this problem, such as the ant colony algorithm that simulates ant behavior (XiongZi-Yuan, XuZhen-Hai. An Innovative subarray partitioning method for clutter suppression by space-time adaptive processing based on the ant colony optimization [J]. IETradarsonarandnavigation,8(9),2014), particle swarm optimization algorithm for simulating bird behavior (FanQin-qin, YanXue-feng.Self-adaptiveparticleswarmoptimizationwithmultiplevelocitystrategiesanditsapplicationforp-Xyleneoxidationreactionprocessoptimization[J].systemChemometricsandintelligentlaboratory0,149), 2 Swarm algorithm (RunklerThomasA.Waspswarmoptimizationofthec-meansclusteringmodel[J]Internationaljournalofintelligentsystems,23(3),2008), artificial fishswarm algorithm for simulating fish foraging behavior (NeshatMehdi, SepidnamGhodrat.Artificialfishswarmalgorithm:asurveyofthestate-of-the-art, hybridizationandind [J].Artificialintelligentreview,42(4),2014) and the cat group algorithm for simulating the daily behavior of cats (P.M.Pradhan.G.Panda.SolvingMulti-ObjectiveProblemsUsingCatSwarmOptimization[J].ExpertSystemswithApplications,3(39),2012), etc. However, these algorithms have their own shortcomings. For example, the ant colony algorithm introduces pheromone, which increases the time complexity of the algorithm, requires a long search time, and is prone to stagnation during the search process; In the later stage, it may not be able to jump out of the local optimum; although the cat swarm algorithm can achieve good results in function optimization, it has problems of slow speed and long time in image processing. In order to solve the problems in image object extraction and classification, a method of object extraction and classification based on improved cat group algorithm is invented.

The so-called target object extraction is a technology to obtain the target object of interest from the image according to the characteristics of the target object, and to identify and segment it. Target object extraction is at the bottom of the entire computer vision system and is the basis for various advanced applications such as target detection and tracking. The quality of the target object extraction is directly related to the tracking of the target object and the quality of the whole system. A good target object extraction algorithm should be applicable to various monitored environments, for example: it can adapt to various weather conditions, adapt to changes in light, adapt to the interference caused by the movement of individual objects in the scene, and can handle shadows and occlusions.

Target object classification technology is to classify objects into predefined categories according to their underlying visual features, and it is an important way to realize automatic computer recognition of targets. The actual operation of target object classification technology mainly includes several stages such as image preprocessing, feature extraction, classifier design and learning. Target object classification methods can be roughly divided into two ways: object classification methods based on generative models (such as Bayesian classification) and object classification methods based on discriminative models (such as k-means classification).

In real life, object classification technology is the core content of solving the above-mentioned computer vision, and is applied to all aspects of real life. For example: the visual navigation of autonomous vehicles, which is based on the classification and recognition of the environment of objects; the reading, discrimination and classification of aerial and satellite photos; the specific target recognition of industrial robots' hand-eye systems; the identification of biological characteristics, etc. Of course, one of the most important application examples of object classification technology is network image retrieval. It not only helps the image retrieval system to understand the semantic information of the image well, but also greatly reduces the manual participation process, and provides strong support for improving the accuracy of the image retrieval system. .

The clustering problem of the swarm intelligence algorithm is to find a division that can minimize the sum of the total intra-class dispersion. When the cluster center is determined, the division of clusters can be determined by the nearest neighbor rule. Suppose there is a target feature set, X={X _i , i=1, 2, . . . , n}, where X _i is an n-dimensional feature vector, and n is the number of feature vectors in X. The purpose of clustering is to find an _{optimal partition C = {} C ₁ , C ₂ , ... C _K , so that the total intra-class dispersion sum reaches J _c minimum, as shown in the following formula:

${\mathrm{<span\; class="notranslate">\; J</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

Among them, C _j is the center of the _jth cluster, d(X _i , C _j ) is the square of the Euclidean distance between the target feature vector Xi and the cluster center C _j , which can be expressed as follows:

d(X _i , C _j )=|X _i -C _j | ²

When the cluster center is determined, the division of clusters can be determined by the nearest neighbor rule. That is, for Xi, if the cluster center C _j of the _jth class satisfies the following formula, then Xi belongs to class _j :

$\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; )</span>$

Some terms:

Target object classification: It is an image processing method that distinguishes different types of target objects according to the different characteristics reflected in the image information. It uses computer to quantitatively analyze the image, and classifies the image or each target or area in the image into one of several categories to replace the human visual interpretation method.

Target object extraction: It is a technique to separate the target object of interest from the background object in the image according to the characteristics of the target object.

Coding of cats: Coding of cats is a formal representation of the solution of a problem. In the cat swarm algorithm, cats are a feasible solution to the problem, and the attributes of each cat include flag bits of speed, position, fitness and behavior mode.

Fitness: Fitness is the degree of adaptation of the individual to the environment, and the evaluation of the individual in the problem that the individual seeks.

Memory pool: The memory pool is the space where the cat copies its own location. The size of the memory pool represents the number of places the cat can search.

Grouping rate: The grouping rate is the quantitative relationship between the two modes of the cat group, and it is the proportion of the cats in the tracking mode to the entire cat group. A small number of cats are in stalking mode and a large number of cats are in hunting mode.

Search mode: In the search mode, the cat copies its own position and puts it in the memory pool. Through the mutation operator, the copied copy in the memory pool is changed, and then the fitness value of the copy is calculated, and the position with the highest fitness value is selected as the next position. step position.

Tracking mode: In tracking mode, the cat will track the "extreme" of the optimal solution found by the entire group of cats to update its own speed and position, so that it will move towards the position of the optimal solution.

Mutation operator: The mutation operator is a local search operation. Each cat is copied and mutated to generate candidate solutions in the neighborhood, and the optimal solution is found in the neighborhood, which completes the mutation operator.

Selection operator: The selection operator is mainly to generate a new position from a copy of the cat's own position in the search mode, put it in the memory pool, and then select the candidate solution with the highest fitness from the memory pool as the new position to replace.

Contents of the invention

The object of the present invention is to provide a method of target extraction and classification based on the improved cat group algorithm, which uses the improved cat group algorithm to perform calculations according to the characteristics of the target object in the image, so as to achieve accurate classification of the target object. The advantages of the invention are: 1) Compared with the original cat group algorithm, the speed is improved and the time is shorter; 2) The target object can be accurately classified.

In order to achieve the above object, the present invention adopts following technical scheme:

(1) Input the original image to be processed;

(2) If the original image is noisy, the image should be preprocessed first. First select an area that is as featureless as the background, estimate the noise model and parameters, and then select the appropriate filter for filtering and denoising according to the noise model. If it is salt and pepper noise, use median filtering; if it is Gaussian or uniform noise, Then use the mean value filter; if it is periodic noise, use frequency domain filtering. If there is no noise or blur, you can skip to the next step;

(3) Segment the preprocessed image threshold into a binary image, and then segment and mark the target image of interest;

(4) Select and calculate four features of the target image: fineness ratio, eccentricity, degree of reliability of the region and degree of expansion of the region to form a new feature vector;

(5) Improve the traditional cat group algorithm to improve the classification speed and accuracy. The improvement of the cat group algorithm is as follows:

① In order to maintain a balance between local search and global search, swarm intelligence algorithms usually use an increasing linear inertia weight, such as particle swarm optimization. It is well known that a large inertial weight is beneficial to global search and a small inertial weight is beneficial to local search, so these algorithms have strong local search ability in the early stage of iteration. However, if the optimum is not found early in the algorithm, it is easy to get stuck in a local optimum. Because as the inertial weight becomes larger, the global search ability becomes stronger. So in order to solve this deficiency, we add a non-linear decreasing inertia weight to the speed update formula of the cat swarm algorithm, as shown in the following formula:

$\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; )</span>\frac{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{{{\mathrm{<span\; class="notranslate">\; iter</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}^{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}}$

Here W _max and W _min are the maximum and minimum values of the inertia weight, t is the number of iterations, iter _O is the critical value, when the number of iterations is iter _O , W(t) is equal to W _max . λ is a constant. The above formula shows that the coefficient will be adaptively decreased nonlinearly;

②In the original cat swarm algorithm, C(t) is the acceleration coefficient in the speed update formula, usually a constant. Here we also let it perform adaptive updates through the following formula:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; iter</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; iter</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}}$

Where t is the number of iterations, iter _max is the maximum number of iterations, and C _i is the initial acceleration coefficient, which is a constant;

③Through the two parameters of the above two parts, the speed update formula of the tracking mode becomes as follows:

V _K,d (t+1)＝W(t)*V _k,d (t)+r ₁ *C(t)*(X _best,d (t)-X _k,d (t))

X _K,d (t+1)=X _K,d (t)+V _K,d (t+1)

V _Kd (t+1) represents the speed value of the k-th cat after the update, X _{best, d} (t) represents the position of the cat with the highest fitness; X _kd (t) refers to the position of the k-th cat , C(t) is a constant, r ₁ is a random number between [0,1]. It can be seen from the above formula that the moving direction of the cat is determined by two parts: its own original speed V _kd (t), the best distance X _{best experienced by the cat group, d} (t)-X _{k, d} (t), The values are determined by the dynamic inertia weight, the acceleration coefficient C(t), and the random number r ₁ respectively;

(6) Use the new feature vector as the input feature library, as the input part of the above improved cat swarm algorithm, and use the new cat swarm algorithm to classify the target object. The detailed steps are as follows:

① Encode the cat, use the new feature vector as the speed of the cat, set the grouping rate, gene change range and memory pool size, etc.;

② Initialize the position and speed of the cat, the maximum number of iterations, etc.; the nearest neighbor clustering, calculate the fitness value according to the new clustering center;

③Let some cats be in the search mode and the other cats in the tracking mode; according to the grouping rate, randomly set the cats in the search mode and the cats in the tracking mode in the cat group, the cat with the flag 0 executes the search mode, and the cat with the flag 1 executes tracking mode;

④ The cat needs to find the global optimal position, update the speed and position of the cat according to the position formula and the improved speed update formula, and approach the optimal solution;

⑤ Copy j copies of each cat's own position, and perform a mutation operator on the copies, and change their positions according to the position formula. Calculate the fitness value of the copy after the position is updated, and select the position with the highest fitness value as the next position for the cat to move;

⑥According to the code of the cluster center of the cat, determine the cluster division of the sample according to the nearest neighbor method, calculate the new cluster center, update the fitness value of the cat, find and record the current optimal solution;

⑦ If the algorithm meets the end condition, then end the algorithm and output the global optimal solution, otherwise transfer to step ③. Finally, the classification label of each target object is returned, and the corresponding result is output.

Description of drawings

Fig. 1 is a schematic diagram of the present invention;

Fig. 2 is one of the original images of the embodiment of the present invention;

Fig. 3 is the image target marking figure of the embodiment of the present invention;

Fig. 4 is the image target extraction and classification result 1 of the embodiment of the present invention;

Fig. 5 is the image target extraction and classification result 2 of the embodiment of the present invention;

Fig. 6 is the image target extraction and classification result 3 of the embodiment of the present invention;

FIG. 7 is a time performance comparison chart of four different algorithms according to an embodiment of the present invention.

detailed description

This embodiment is realized on the PC provided by the Institute of Intelligent Information of Hunan University of Technology, the processor of this machine is Intel(R) Pentium(R) CPUG2030 3.00GHz4GB memory, the operating system used is windows7, and the simulation software used is MATLAB2014a.

The relevant steps of the present invention are described in detail in conjunction with accompanying drawing 1 and embodiment, and the present invention is generally divided into following several parts:

(1) Input image. Input the original image to be processed, as shown in Figure 2;

(2) Image preprocessing. If the original image is noisy, the image should be preprocessed first, such as image denoising and other operations, so as to remove the interfering noise in the image. First, an area that is as featureless as the background should be selected, the noise model and parameters should be estimated, and then an appropriate filter should be selected according to the noise model for filtering and denoising. If it is salt and pepper noise, use median filter; if it is Gaussian and uniform noise, use mean filter; if it is periodic noise, use frequency domain filter. If there is no noise or blur, you can skip to the next step;

(3) Target object extraction. First, the preprocessed image threshold is segmented into binary images, and then the target image of interest is segmented and marked for selection, as shown in Figure 3, in order to carry out the next step;

(4) Feature selection and calculation. According to the target image extracted in the previous step, select and calculate the four features of the target image, such as fineness ratio, eccentricity, solidity and degree, to form a new feature vector;

(5) Classification of target objects. Input the features calculated in the previous step into the improved cat group algorithm, and classify the target objects according to the following steps:

①Initialize parameters and establish initial population: encode the cat, use the new feature vector as the speed of the cat, set the grouping rate, gene change range and memory pool size, etc. Initialize the position and velocity of the cat, the maximum number of iterations, etc.;

② Calculate the fitness value: perform nearest neighbor clustering on the target features, and calculate the fitness value according to the new cluster center;

③Set the grouping rate: let some cats be in the search mode, and the other part of the cats are in the tracking mode;

④ Judgment mode status: According to the grouping rate, randomly set the cats in the search mode and the cats in the tracking mode in the cat group, the cat with the flag bit 0 executes the search mode, and the cat with the flag bit 1 executes the tracking mode;

⑤Tracking mode: the cat needs to find the global optimal position, update the speed and position of the cat according to the new improved speed update formula, and approach the optimal solution;

⑥ Copy j copies of each cat's own position, and perform a mutation operator on the copies, and change their positions according to the formula. Calculate the fitness value of the copy after the position is updated, and select the position with the highest fitness value as the next position for the cat to move;

⑦ Calculate the fitness value and retain the optimal solution: According to the code of the cluster center of the cat, determine the cluster division of the sample according to the nearest neighbor method, calculate the new cluster center, update the fitness value of the cat, find and record the current most Excellent solution;

⑧Judging whether the end condition is satisfied: if the algorithm meets the end condition, then end the algorithm and output the global optimal solution, that is, the classification result of the target object, as shown in Figure 4, Figure 5, and Figure 6; otherwise, transfer to step ③.

In the present invention, in order to analyze the time performance of the algorithm, we apply the algorithm of the present invention (ICSO), the cat swarm algorithm (CSO), the particle swarm algorithm (PSO) and the ant colony algorithm (ASO) to the original image respectively. Conduct one hundred experiments respectively, record and compare the experimental results. The specific experimental results are shown in Figure 7, which is a graph comparing the number of times and time. The horizontal axis represents the number of times the algorithm runs, and the maximum number of iterations is 100; the vertical axis represents the time result, and the unit is seconds (s). It can be seen from the results that the top cyan time axis is the ant colony algorithm (ASO), then the green cat swarm algorithm (CSO), followed by the blue particle swarm algorithm (PSO), and the bottom is the red Ben Invent algorithms. It shows that the time of the ant colony algorithm is the longest, and the time of the improved algorithm of the present invention is the shortest, so this algorithm is superior to other algorithms in terms of time performance, and is also improved and improved compared with the original algorithm.

Claims (3)

1. A method based on the target extraction and classification of the improved cat group algorithm, characterized in that: The traditional cat swarm algorithm is improved: firstly, a nonlinear inertia parameter is introduced into the speed update formula in the tracking mode of the traditional cat swarm algorithm; secondly, a linear autocorrelation coefficient is added in order to expand the search space; in addition, In order to increase the search speed and effect, the optimal retention strategy is used in the algorithm; finally, the improved algorithm is used to extract and classify the target objects. 2. a kind of method based on the target extraction of improved cat group algorithm and classification according to claim 1, traditional cat group algorithm is improved, it is characterized in that: ①In order to maintain a balance between local search and global search, an increasing linear inertia weight is used for adjustment. A large inertia weight is beneficial to global search and a small inertia weight is beneficial to local search, because as the inertia weight becomes larger and larger , the global search ability is getting stronger and stronger. Therefore, we add a non-linear decreasing inertia weight to the speed update formula of the cat swarm algorithm. The formula is as follows Here W _max and W _min are the maximum and minimum values of the inertia weight, t is the number of iterations, iter ₀ is the critical value, when the number of iterations is iter ₀ , w(t) is equal to W _max , λ is a constant, above The formula shows that the coefficient will be adaptively decreased nonlinearly; ②In the original cat swarm algorithm, C(t) is the acceleration coefficient in the speed update formula, which is usually a constant. In this patent, we also let it update adaptively through a formula: Where t is the number of iterations, iter _max is the maximum number of iterations, and C _s is the initial acceleration coefficient, which is a constant; ③Through the two parameters of the above two parts, the speed update formula of the tracking mode becomes as follows: V _K,d (t+1)=W(t)*V _k,d (t)+r ₁ *C(t)*(X _best,d (t)-X _k,d (t)) V _{K, d} (t+1) represents the speed value of the k-th cat after the update, X _{best, d} (t) represents the position of the cat with the highest fitness; X _{k, d} (t) refers to the k-th cat The position of the cat, C(t) is a constant, r ₁ is a random number between [0,1]. 3. a kind of method based on the target extraction of improved cat group algorithm and classification according to claim 1, the method technical scheme that target is extracted and classification is as follows: (1) Input the original image to be processed; (2) If the original image is noisy, first preprocess the image: first select an area that is as featureless as the background, estimate the noise model and parameters, and then select the appropriate filter for filtering and denoising according to the noise model. If If it is salt and pepper noise, use the median filter; if it is Gaussian or uniform noise, use the mean filter; if it is periodic noise, use frequency domain filtering; if there is no noise or blur, you can skip the next step; (3) Segment the preprocessed image threshold into a binary image, and then segment and mark the target image of interest; (4) Select and calculate four features of the target image: fineness ratio, eccentricity, degree of reliability of the region and degree of expansion of the region to form a new feature vector; (5) Improve the traditional cat group algorithm to improve the classification speed and accuracy. The improvement of the cat group algorithm is as follows: ①Swarm intelligence algorithm In order to maintain a balance between local search and global search, a decreasing linear inertia weight is used in this patent, because a large inertia weight is beneficial to global search and a small inertia weight is beneficial to local search. Therefore, We add a non-linear decreasing inertia weight to the speed update formula of the cat swarm algorithm, as shown in the following formula: Here W _max and W _min are the maximum and minimum values of the inertia weight, t is the number of iterations, iter ₀ is the critical value, when the number of iterations is iter ₀ , w(t) is equal to W _max , and λ is a constant; ②In addition, in the original cat swarm algorithm, C(t) is the acceleration coefficient in the speed update formula, which is usually a constant. Here we also let it perform adaptive update through the following formula: Where t is the number of iterations, iter _max is the maximum number of iterations, and C _s is the initial acceleration coefficient, which is a constant; ③Through the two parameters of the above two parts, the speed update formula of the tracking mode becomes as follows: V _K,d (t+1)=W(t)*V _k,d (t)+r ₁ *C(t)*(X _best,d (t)-X _k,d (t)) X _K,d (t+1)=X _K,d (t)+V _K,d (t+1) V _{K, d} (t+1) represents the speed value of the k-th cat after the update, X _{best, d} (t) represents the position of the cat with the highest fitness; X _{k, d} (t) refers to the k-th cat The position of the cat, C(t) is a constant, r ₁ is a random number between [0,1]; from the above formula, it can be seen that the moving direction of the cat is determined by two parts: its original speed V _{K , d} (t), and the best distance X _{best experienced by cats, d} (t)-V _{K, d} (t), are determined by the dynamic inertia weight, acceleration coefficient C (t) and random number r ₁ respectively value; (6) Use the new feature vector as the input feature library, as the input part of the above improved cat swarm algorithm, and use the new cat swarm algorithm to classify the target object. The detailed steps are as follows: ① Encode the cat, use the new feature vector as the speed of the cat, set the grouping rate, gene change range and memory pool size, etc.; ② Initialize the position and speed of the cat, the maximum number of iterations, etc.; the nearest neighbor clustering, calculate the fitness value according to the new clustering center; ③Let some cats be in the search mode and the other cats in the tracking mode; according to the grouping rate, randomly set the cats in the search mode and the cats in the tracking mode in the cat group, the cat with the flag 0 executes the search mode, and the cat with the flag 1 executes tracking mode; ④ The cat needs to find the global optimal position, update the speed and position of the cat according to the position formula and the improved speed update formula, and approach the optimal solution; ⑤ Copy j copies of each cat's own position, and perform a mutation operator on the copies, change their positions according to the position formula, calculate the fitness value of the copy after the position update, and select the position with the highest fitness value as the cat movement the next position of ⑥According to the code of the cluster center of the cat, determine the cluster division of the sample according to the nearest neighbor method, calculate the new cluster center, update the fitness value of the cat, find and record the current optimal solution; ⑦ If the algorithm meets the end condition, then end the algorithm, and finally return the classification label of each target object, and output the corresponding result; otherwise, transfer to step ③.