Academic Journal of Science and Technology 
ISSN: 2771-3032 | Vol. 9, No. 2, 2024 

230 

Tool	Damage	Detection	Method	Based	on	Improved	
Threshold	Segmentation	Algorithm	 	
Lunwen Peng1, Shenghu Pan2 

1Department of Mechatronics Engineering, Southwest Petroleum University, Chengdu 610500, China 
2Oil and gas equipment technology Sharing and Service Platform of Sichuan Province 610500, China 

Abstract: In order to solve the problem that the current tool wear defects are difficult to be collected by the visual inspection 
system, an Otsu threshold segmentation algorithm based on particle swarm optimization is proposed.to detect tool wear The 
algorithm improved update strategy for inertia coefficients which effectively expanding the search scope of the algorithm and 
shortens the running time of the algorithm. By adding a perturbation equation to the particle swarm, solved the problem of 
traditional particle swarm optimization algorithms easily falling into local optima. Finally, an experimental platform is built to 
verify the effectiveness of the algorithm. This inspection method can achieve the identification of tool damage areas and the 
measurement of tool damage amount, and has advantages such as high recognition accuracy and fast running speed compared to 
traditional Otsu algorithm, Canny algorithm ,local threshold segmentation and so on. The research results have certain reference 
value for the actual tool defect detection system. 

Keywords: Tool wear; Visual detection; Otsu: PSO. 

1. Introduction
At present, cutting is still one of the important machining 

processes, and the tool as the core tool in cutting, the quality 
of the tool will affect the machining accuracy and quality of 
the machine tool. In actual processing, it is usually 
determined by worker experience whether the tool has 
reached scrap conditions. This method will result in a 
considerable number of tools that have not yet reached the 
scrapping conditions to be replaced. If the tool has been 
excessively worn and has not been replaced, it will 
significantly reduce the dimensional accuracy of the parts and 
even cause safety accidents. Therefore, in order to accurately 
measure the wear of the tool, it is necessary to detect the tool 
wear area. 

The method of detecting and measuring tool state can be 
divided into direct measurement method and indirect 
measurement method. Indirect measurement method is to 
detect the characteristic changes of cutting force, sound, 
vibration, workpiece surface texture and other related 
physical quantities caused by tool wear through sensors, so as 
to measure the tool indirectly. 

The indirect measurement method has some disadvantages: 
First, to obtain the characteristics of tool wear from the signal 
requires the operator to know the relevant domain knowledge; 
Secondly, sensor-based tool detection devices are often only 
applicable to a certain type or type of tool, and the 
generalization is poor and the signal detected by the sensor is 
doped with a large number of interference signals, which will 
affect the test results [1]. 

The direct method is to detect the defect directly on the 
image of the tool. Ye Zukun et al. [2] proposed a damage 
detection method based on local threshold segmentation. The 
whole image is divided into many small pixel blocks, the 
segmentation threshold of each pixel block is obtained, and 
the whole pixel is scanned according to the maximum 
segmentation threshold, and the tool damage location is 
obtained. However, due to the complex surface texture of the 
tool, the acquired tool image contains a lot of noise, which is 

easy to select the noise as the segmentation threshold and lose 
the damage location information. Deng Xiaopeng et al. [3] 
combined Otsu algorithm and region growth method to 
propose an adaptive regional growth defect detection method, 
which can automatically locate the tool wear area and identify 
the size of tool damage. However, Otsu algorithm is 
susceptible to the influence of tool texture and reflection, 
resulting in difficulty in locating the tool defect location. 
Zhou Junjie [4] et al. proposed a method of edge detection 
followed by pixel detection. Firstly, Canny operator is used to 
rough position the image, then the sub-pixel edge detection 
algorithm of Zernike moment is used to improve the 
measurement accuracy, and finally the pixel points are 
simulated to complete the measurement. However, this 
method is very sensitive to grayscale fluctuations, and it is 
easy to identify the inner texture of the tool as the boundary 
information, which affects the recognition accuracy, and the 
edge contour discontinuity is easy to occur. Li Shanshan et al. 
[5] obtained the tool defect region through the region growth
method, and obtained the tool characteristic value through the
minimum external matrix. However, it is difficult to calculate
the damage amount accurately because the gray value of the
tool damage surface fluctuates greatly. Liu Jianchun et al. [6]
proposed an adaptive threshold segmentation algorithm,
which combined collinear contours to fit the edge contour and
extract the defect area. However, due to the small defect area
of the tool, it is easy to identify the non-wear area as the wear
area.

2. Tool image damage detection
method

2.1. Image preprocessing 
2.1.1. Image graying 

RGB model is a color model defined according to human 
eye perception. RGB model converts color images into 8-bit 
gray images according to different proportions of three-
channel pixel values. According to experiments, this ratio 
column coefficient has the best effect in processing color 



231 

images [7]. R, G and B respectively represent the three color 
components of red, green and blue. In general,, and the 
formula is as follows: 

0.299 0.587 0.114Gray R G B         (1) 

The image is imported into the software platform for gray 
level conversion, and the original image is taken on the 
experimental platform. The image before conversion is shown 
in Figure a), and the effect of the tool after gray level 
conversion is shown in Figure b).  

a) Tool raw image b) Tool gray image
Figure 1. Image graying 

2.1.2. Sub heading 
In the process of image acquisition, transmission and 

compression, the camera will be affected by light, which will 
lead to the acquisition of images containing noise, noise will 
lead to the loss of key information in the image, and noise 
may bring certain interference to the identification of defects. 
The function of image filtering is to eliminate the noise in the 
image and dilute the irrelevant information, retain the 
effective information in the image, and improve the signal-to-
noise ratio So it is necessary to use the mean filtering 
algorithm to eliminate the noise and smooth the image  

The mean filtering algorithm is a method to smooth the 
image contour. The algorithm sets a window (2a+1)(2b+1), 
sets each pixel on the image as the center point of the window 
for traversal, and assigns a value to the center pixel with the 
median value of the pixel proximity. The average value of 
each window pixel is: 

( , ) ( , )
1

'
(2 1)(2 1)

a b

r h r i h j

i a j b

g g
a b

 

 


      (2) 

Where, g (r,h) is the average gray value of each pixel in the 
window, and g '(R-I, H-j) is the gray value of each pixel in the 
window.  

2.2. Otsu threshold segmentation method. 
Threshold segmentation method is a common image 

segmentation method, the main purpose of threshold 
segmentation is to classify each pixel in the image through the 
difference of pixel gray value, and realize the distinction 
between background and target, so as to extract the interesting 
target from the image. 

Suppose an image has L gray levels, ni (0< i< L-1) is 
the number of pixels in different gray levels, then the total 
number of pixels N of the image is: 

1

0

L

i
i

N n




        (3) 

Normalize the grayscale histogram: 

1

0

1
L

i
i

p




             (4) 

Then the probability that the gray value of the pixel in the 
figure is i is: 

i
i

n
p

N
   (5) 

Select a threshold value K, and divide pixels into two 
intervals X0 and X1 according to different gray levels. X0 is 
the pixel with gray level [0, k-1], and X1 is the pixel with gray 
level [K, L-1]. Then the probability of the pixel falling into 
the two intervals is: 

1 1

0 1 0
0

, 1
K L

i i
i i K

x p x p x
 

 

          (6) 

The mean gray values of X1 and X2 are: 

1 1

0 1
00 1

1 1
,

K L

i i
i i K

ip ip
x x

 
 

 

               (7) 

The variance between X0 and X1 classes is: 

2

0

0

2 2
0 1 1

1 1 0

( )x x

x x

    

 

   

 

（ ）

（ ）
  (8) 

The classical Otsu segmentation method is to iterate the 
value of threshold K in the gray level in the range of [0, L-1] 
until the optimal segmentation threshold is obtained. In this 
paper, the Improved Particle Swarm Optmization algorithm is 
used to optimize the optimal threshold of Otsu algorithm, so 
as to improve the convergence speed and accuracy of the 
algorithm. 

2.3. PSO algorithm. 
Particle Swarm Optimization (PSO) abstracts a flock of 

birds into particles without volume and mass, takes the 
position of individual birds or food as the optimal solution, 
and the particles fly at a certain speed in the solution space, 
using the information of the group to interact with the 
information of the optimal individual. In order to adjust its 
flight speed, the entire particle swarm converges towards the 
optimal individual under the condition of ensuring population 
diversity [10]. The algorithm steps are as follows: 

First, let particle swarm size be N, search space dimension 
be D, position vector Xi={x1, x2, x3, xiD}, velocity vector 
Vi{V1, V2, V3, ViD,}. i is the particle number, the position 
vector of the particle represents the position of the solution in 
the solution space, and the velocity vector represents the 
direction and rate of the particle moving in the solution space. 
The particle swarm finds the optimal solution through 
continuous iteration, and the particle iteration formula is as 
follows: 

( 1) ( ) ( 1)iD iD iDX t X t V t          (9) 

Where, t is the number of iterations, ω is the inertia weight 
of the particle, representing the influence of the previous 



 

232 

iteration speed on the current speed. c1 and c2 are 
acceleration factors, c1 controls the learning of individual 
particles, and c2 controls the mutual learning between 
particles and particle swarm. Pbest is the optimal solution 
reached by the individual, Gbest is the optimal solution 
reached by the group. 

The determination of inertia weight is a very important part 
of particle swarm optimization algorithm, which will directly 
affect the convergence speed and accuracy of the algorithm. 
The value of ω is high and the search range of the algorithm 
is large. The value small algorithm of ω converges quickly. 
The traditional PSO algorithm has some defects, such as low 
precision and slow convergence speed, because its inertia 
weight ω is constant. At present, most studies [11][12] modify 
the parameter part of the algorithm, such as improving the 
decline strategy of inertia weight, such as linear decline 
strategy of inertia weight, that is, the inertia weight decreases 
with the increase of the number of iterations. 

The updated formula is: 
 

max max min
m

M
      （ ）          (10) 

 
ωmax, ωmin are the maximum and minimum inertial 

weights. m is the current number of iterations, and M is the 
total number of iterations. The adaptability of this update 
strategy is wide, and the convergence of the algorithm is 
improved on the basis of guaranteeing speed and accuracy, 
but there are some defects in this algorithm. In the early stage 
of the algorithm, more emphasis should be placed on search 
to improve the optimization ability of the algorithm, but the 
linear decline strategy of inertia weight will lead to the weak 
search ability of the particle swarm in the early stage and can 
not find the global optimal solution. In the later stage, the 
algorithm should focus more on convergence, and the more 
the inertia weight linear decline strategy converges, the 
slower the algorithm runs. To sum up, in order to improve the 
convergence speed and accuracy of the algorithm, the inertia 
coefficient cosine decreasing strategy is adopted to improve 
the optimization ability of particle swarm in the early stage 
and the convergence ability of particle swarm in the later 
stage. The updated formula is as follows: 

 

max min

max min

1
[cos( ) ( )

2
( )]

k

m

M
   

 

   


     (11) 

 
A large number of studies have found that when the 

maximum inertia weight value ωmax is 0.9 and the minimum 
inertia weight value ωmin is 0.4[13], the algorithm 
performance will be greatly improved. Set the maximum 
number of iteration M of the algorithm to 100 and bring it into 
the formula for calculation. 

 
Figure 2. Inertia weight coefficient comparison 

 

he results are shown in Figure 2. It can be seen that in the 
early stage of iteration, the value of the inertia coefficient ω 
is large, and the particle swarm has a strong ability to conduct 
global search, which is conducive to the group's global 
learning and avoids the algorithm falling into the local 
optimal solution. In the late iteration, the inertia weight value 
ω is smaller, the particle local search ability is stronger, and 
the search speed is faster, which is conducive to improving 
the search efficiency of PSO algorithm. 

However, the search result of the algorithm may still be the 
local optimal solution. However, it can be confirmed that 
when the particle swarm converges towards the global 
optimal solution, the individual optimal value in the next 
iteration must be higher than the individual optimal value in 
the last iteration. Therefore, judging the size of the individual 
optimal position response function beforeand after the 
iteration can determine whether a single particle is evolving 
[14], that is: 

 

 
( , ) ( , 1)

,
( , ) ( , 1)

1         

0        

xiD t xiD t
i t

xiD t xiD t
A

f f

f f
 

 






  

＞
      (12) 

 
Where, it represents whether the I-th particle in the t 

iteration is evolving towards the particle individual 
optimization. 

 

 ,

1
( )

N

i t

t
s t

A
P

N

               (13) 

 
Where, N represents the total number of particles, and Ps(t) 

is a value between [0,1], the size of which represents the 
number of individuals in the whole particle swarm evolving 
toward the individual optimal position. If Ps(t) is large, most 
of the particles in the population have not reached the global 
optimal position; if Ps(t) is small, it is considered that most of 
the particles in the population have reached the global optimal 
position. If we set the termination condition as Ps(t) < 0.05, it 
indicates that the particle swarm may reach the optimal 
position of the population. At this time, the algorithm 
continues to perform three iterations. If the fitness value of 
the particle swarm after iteration is still higher than that of the 
previous iteration, it is determined that the particle swarm has 
reached the optimal position of the population for the first 
time. However, the particle swarm may still be a local optimal 

20 40 60 80 100
0.4

0.5

0.6

0.7

0.8

0.9

W
ei

gh
ts

Number of iterations

 Cosine decrement
 Lineardecrement



 

233 

solution at this time. After recording the current optimal 
position, a perturbation [15][16] is added to the particle 
swarm to make the particle swarm abandon the current 
optimal solution and find the global optimal solution again. 
After adding the disturbance, the updated equation is: 

 

1 2,
i

i i k

f f

f f fs e s e




            (14) 

 

1

2

( 1) ( ) ()[ ( )]

rand()[ ( )]
iD iD iD iD

iD iD

x t x t s rand Pbest x t

s Gbest X t

   
 

(15) 

 
Where, s1 and s2 are the updated acceleration factors, are 

the fitness value of the i th particle, are the global optimal 
fitness value, and are the minimum value of non-zero. The 
difference between equation (16) and equation (9) is that by 
introducing the acceleration factor update equation, 
perturbations of different intensity are added to the particle 
swarm, so that the particles jump out of the current search area, 
search again, and get closer to the global optimal position, so 
as to prevent the particle swarm from falling into the local 
optimal solution. 

2.4. Improved Otsu threshold segmentation 
process of particle swarm optimization 
algorithm.  

The algorithm flow is shown in Figure 4. The 
implementation steps of the algorithm are as follows: 

1) The tool image is preprocessed to obtain the gray level 
histogram of the image, and then the gray level histogram is 
normalized. 

2) The inter-class variance formula (8) is derived according 
to equation (6) and equation (7). In order to find the maximum 
segmentation threshold for inter-class variance, the improved 
and optimized PSO algorithm is adopted. 

3) Set the particle swarm size to N, randomly initialize the 
particle position and initial velocity, and set the maximum 
inertia weight value ωmax, minimum inertia weight value 
ωmin, acceleration coefficient c1,c2, maximum number of 
iterations, etc. 

4) Use formula (8) as fitness function to calculate the 
fitness value of particles, and update the individual optimal 
position Pbest and global optimal position Gbest for each 
particle according to the fitness value. 

5) Use equations (9) and (10) to calculate the position and 
velocity of the particle 

Update: In the iterative process, the inertia coefficient of 
the particle swarm is reduced by the cosine according to 
equation (12). 

6) Iterate for three consecutive times to determine f (xm+) 
&gt; Whether f(xm) is true (the state function fitness value of 
this iteration is greater than the state function fitness value of 
the last iteration) If yes, jump to Step 5), otherwise return to 
step 4). 

7) Use formula (16) to add perturbations to the particle 
swarm. 

8) Three consecutive iterations to determine f(xm+1)&gt; 
If yes, go to Step 7. If no, go to Step 3. 

9) Calculate the updated particle fitness value and record 
the result 

To the individual optimal position and the global optimal 

position. Then, we continue to use formula (9) and (10) to 
update the particle position and velocity. 

10) If the number of continuous invariable particle swarm 
optimal adaptation values reaches the specified value or the 
number of algorithm iterations reaches the set value. Then the 
global optimal position Gbest is output as the optimal 
segmentation threshold K. 

11) Set K as the segmentation threshold for image 
segmentation. 

3. Experimental Results and Analysis 
In order to verify the effectiveness of the application of the 

improved particle swarm optimization Otsu algorithm in the 
process of tool defect detection, this paper built an image 
acquisition platform, as shown in FIG. 4, which mainly 
consists of 5 million pixel CCD camera, 18FA0828-6MP 
industrial lens, dot matrix light source, base and camera 
support [18]. The base is placed on a horizontal table, and the 
base is provided with a dot matrix light source to provide a 
good image acquisition environment for the camera. The 
camera bracket is vertically fixed on the base, the industrial 
camera is fixed through the beam bracket and is horizontal 
with the vertical bracket, and the image detection table is 
vertically fixed with the vertical bracket. 

 

 
Figure 3. Image acquisition device 

 

Before preparing for testing, it is first necessary to clean the 
tool. Then fix the tool on the image detection table, adjust the 
distance between the camera and the tool by adjusting the 
beam support to get a clear image of the tool damage. Finally, 
the CCD industrial camera is used to collect the information 
of the tool back tool face and then transmit it to PC for the 
next step of image processing. 

PC configuration: 
Intel(R)Core(TM)i5104004,8GRAM,MATLAB
R2022a. 

In order to verify the effectiveness of the tool detection 
algorithm proposed in this paper, 4 tools with different wear 
degrees were selected as samples for experiments. The real 
wear of the tool was measured by the microscope. At the same 
time, the algorithm in this paper was used to carry out 
repeated experiments on four tools, record the experimental 
results and calculate the average accuracy. The average 
accuracy rate is calculated as follows: 

1

1 | |n
i i

ii

x y

n y





 （1- ）100%

 
Where, is the average accuracy, is the true value, is the 

measured value, and n is the number of experiments. The 
measured values were selected to calculate the average 
damage width of the tool surface. 

Figure 3 shows the recognition effect of the improved 
PSO+Otsu algorithm for four tools with different wear 



 

234 

degrees and different lighting conditions. It can be seen that 
in different lighting environments, the algorithm has a good 
effect on noise processing, and can accurately and effectively 
extract the tool damage area. #1 is the first tool, #2 is the 
second tool, #3 is the third tool, and #4 is the fourth tool. The 

figure shows the average result of the damage amount of the 
four tools by the algorithm in this paper. (a) is the tool damage 
length, and (b) is the average width of the tool damage area. 
(c) is the tool damage area of the knife.  

 

 
Figure 4. Effectiveness of the proposed algorithm in identifying different tool damage 

 

 
Figure 5. Different algorithms detect tool defect areas 

 

 
Figure 6. The PSO-Otsu algorithm was improved to detect the result of tool damage 

 

In order to verify the superiority of the proposed algorithm, 
Otsu method, Canny edge detection algorithm, adaptive 
threshold segmentation method and improved PSO+Otsu 
segmentation algorithm are used respectively. Taking the 
vertical turning tool as an example, the same turning tool is 
tested five times, and the average running time, iteration times 
and accuracy of each algorithm are recorded. The quantitative 
analysis of the detection result data shows that the traditional 
threshold segmentation method is susceptible to the influence 

of illumination, which makes it difficult to locate the tool 
defect location, and even unable to identify the damaged area 
in severe cases, as shown in Figure 11 (b). However, the 
traditional edge detection method is prone to problems such 
as discontinuity of edge  

contour, as shown in Figure 11 (c), and cannot form a 
closed area to calculate the tool damage area. The tool damage 
detection algorithm used in this paper can automatically and 
effectively identify the tool damage area, and the recognition 



 

235 

effect is outstanding, the recognition efficiency is faster, and the anti-interference is good. 
 

Table 1. Four Scheme comparing 

Algorithm Run time /ms Average accuracy Number of iterations 
Otsu method 577 - 202 

Canny method 496 74.85 - 
Adaptive threshold method 456 85.84 189 

PSO+Otsu method 189 96.16 22 
 

As can be seen from FIG. 5 and FIG. 6, the tool damage 
detection method adopted in this paper can automatically and 
effectively identify the size of the complete tool damage area, 
with high accuracy, clear identification area and good effect. 
According to Figure 6 (b) and Table 1, although the traditional 
Otsu segmentation method has many iterations and long 
operation time in searching for the optimal threshold of tool 
segmentation, it still fails to identify the tool damage area. As 
shown in Figure 6 (c) and Table 2, defects such as edge loss 
appear in the image of Canny edge detection method, and the 
algorithm has a long running time, low accuracy and poor 
effect. According to FIG. 6 (d) and FIG. 6 (e) in combination 
with Table 1, it can be seen that both of them can clearly 
obtain the tool damage contour and the image is clear, and the 
improved PSO+Otsu algorithm can reach 96.16% accuracy in 
detecting the average width of the tool surface after the tool, 
which is at least 9.81% higher than the other six algorithms, 
and also has advantages in terms of running time and number 
of iterations. It can be seen that the improved PSO+Otsu 
algorithm proposed in this paper has significantly improved 
the accuracy and processing speed of tool image damage 
detection compared with Otsu segmentation method, Canny 
edge detection method, local threshold segmentation method 
and other algorithms. 

4. Summary 
(1) Aiming at the problems of the traditional Otsu 

algorithm, such as large computation amount, long running 
time and easy local optimal solution of PSO algorithm, a 
particle swarm optimization of Otsu threshold segmentation 
tool damage image visual detection algorithm is proposed. 
The algorithm adopts the inertia coefficient cosine decreasing 
strategy to improve the convergence speed of the algorithm, 
and adds the disturbance equation. When the particle swarm 
algorithm falls into the local optimal solution, the disturbance 
is added to the particle swarm to increase the ability of the 
algorithm to jump out of the local optimal solution. 

(2) An experimental detection platform was set up in this 
paper to detect the tool. The experimental results show that 
this method can effectively identify the tool damage area and 
extract the tool damage amount, which has obvious 
advantages over the traditional detection algorithm. It can 
provide strong technical support for in-place detection of tool 
damage. 

(3) All the methods in this paper are implemented on the 
PC side, and the algorithm in this paper is simple and the 
calculation amount is small. Therefore, the method in this 
paper can be transplanted to the embedded platform for code 
adjustment and optimization, so as to apply to more 
application scenarios. 

Acknowledgment 
Sichuan Natural Science Foundation 2022NSFSC2002 

References 
[1] DING Yuhui, Cao Yan, Fu Leijie, BAI Yu. Research status and 

Prospect of Image Processing Technology in Tool Wear 
Detection [J]. Manufacturing Technology & Machine Tool, 
2020,(4): 56-62. 

[2] Ye Zukun, Li Heng, Cha Wenbin, He Yan, Wang Yulin. Tool 
damage visual detection method using local threshold 
segmentation [J]. Journal of Xi 'an Jiaotong University,2021, 
vol. 55 (4): 52-60. 

[3] Deng Xiaopeng, Wang Yan, Hong Yu, Hu Xiaofeng. Micro-
drilling and Milling Tool wear Detection Method using 
adaptive region Growth [J]. Journal of Xi 'an Jiaotong 
University,2021, vol. 55. 

[4] Zhou Junjie, Yu Jianbo. Online Measurement of Machining 
tool wear based on Machine vision [J]. Journal of Shanghai Jiao 
Tong University,2021, vol. 55 (6): 741-749. 

[5] Li Shanshan1, Liu Libing1, Li Li1 et al. Nc Tool Wear State 
Detection Method Based on Region Growth Method [J]. 
Manufacturing Technology & Machine Tool, 2017, (2):132-
135. 

[6] Liu Jianchun, Jiang Junjie, Zou Pilgrimage. Research on Wear 
Detection Method of End Milling Cutter Based on Machine 
Vision [J]. Manufacturing Technology & Machine 
Tool,2020,(1): 136-140. 

[7] Zhang Guihua, Feng Yanbo, Lu Weidong. Gray-scale and 
feature region acquisition of image processing [J]. Journal of 
Qiqihar University: Natural Science Edition, 2007,23(4):49-52. 

[8] Tang Jian. Research on solar multi-band image registration 
method [D]. Kunming University of Science and Technology. 

[9] Lu Yan. Research on image segmentation based on threshold 
algorithm [D]. Chongqing University,2011 

[10] Liu Guihong, Zhao Liang, Sun Jinguang, Wang Wang. An 
improved particle swarm optimization algorithm for Otsu 
image threshold segmentation [J]. Computer Science,2016,(3): 

[11] Dong Hongbin, Li Dongjin, Zhang Xiaoping. A Particle swarm 
optimization algorithm for Dynamic Adjustment of inertia 
weights [J]. Computer Science,2018,45(2):98-102139. (in 
Chinese). 

[12] AO Yongcai, Shi Yibing, Zhang Wei, Li Yanyun. Improved 
particle swarm optimization based on adaptive inertia weights 
[J]. Journal of University of Electronic Science and 
Technology of China,2014,(6): 874-880. 

[13] Cheng Wansheng, Zang Xizhe, Zhao Jie, CAI Hegao. Optics 
and Precision Engineering,2008,(10): 1907-1912. Improved 
PSO inertia factor for Otsu Threshold Search [J]. 

[14] LIU Shenxiao, Wang Xuechun, Chang Chaowen. Otsu image 
segmentation method based on improved particle swarm 
optimization [J]. Computer Science,2013,(8): 293-295. 

[15] Zhang Xinjuan, Lei Xiujuan. Application of improved PSO 
algorithm in two-dimensional optimal threshold image 
segmentation [J]. Computer Engineering and Applications, 
2011,  (26). 



 

236 

[16] Tian Xinghua. Research on improved particle swarm 
optimization based on Chaotic mapping [D]. Qingdao 
University, 2020. 

[17] Xu Zu. Iso 3685-1977 (E) Life test of single-edge turning tools 
[J]. Tool Technology,1979,(4): 55-65. 

[18] Qin Guohua, Yi Xin, Li Yilan, Xie Wenbin. Optical Precision 
Engineering,2014, vol. 22 (12): 3332-3341.