Automatic detection of feather defects using Lie group and fuzzy Fisher criterion for shuttlecock production

doi:10.1016/j.ymssp.2020.106690

Mechanical Systems and Signal Processing

Volume 141, July 2020, 106690

https://doi.org/10.1016/j.ymssp.2020.106690 Get rights and content

Highlights

•
Image enhancement based on fuzzy sets can effectively enhance the boundary of feather quill.
•
Covariance matrix is constructed as features of the defect region.
•
Mean covariance matrix with nonsymmetric structure is proved to have Lie Group structure.
•
The recognition of stain defects can be accomplished by the distribution of color space.
•
The recognition method combining Riemannian metric and FFLD has stable defect recognition ability.

Abstract

Feathers are the main raw material of shuttlecock. The quality of feather is a key factor in shuttlecock production and detection of feather defects is a challenging problem. In this paper, an automatic detection for feather defects was proposed that relies on fuzzy, Lie Group and machine learning theory. First, the feather images were enhanced using fuzzy set. Then, a covariance matrix was constructed as features of the defect region and Riemannian metric was assigned on a Riemannian manifold with a Lie group structure. The mean covariance matrix with nonsymmetric structure was proved to have a Lie group structure. To establish the automatic detection of feather defects, this study introduced Riemann mean and membership degree into the machine learning method of Fisher linear discriminant to implement feature recognition. Experiments were performed to demonstrate the feasibility of features of defect region as a differential manifold.

Introduction

Feather pieces made from natural goose and duck feathers are important materials in the production of shuttlecocks. Sixteen feathers are required to produce a shuttlecock. A high-grade shuttlecock requires additional uniform feathers to ensure its flying quality [1], [2], [3]. The classification index of feather pieces is numerous and complicated. At present, no national unified classification standard is available. The classification standards of shuttlecock manufacturers are set in accordance with industry practices. Fig. 1 shows the definitions of feather root, feather quill, feather leaf, and other parts.

The feathers of ducks and geese may naturally become damaged during the growth process and in the subsequent stage of feather piece processing. Feather defects include wormhole (insect-induced), lack, scar, stain, wrinkle, crease, ripple and overall discoloration. Such feather defects are the key factors that affect quality. Fig. 2 shows a feather piece with defects caused by wormhole, lack, and stain. The quality of feather sorting directly determines the quality of the finished shuttlecock product and ultimately affects the production efficiency and economic benefits of the manufacturer. Therefore, accurately sorting the defective feather pieces is necessary in producing quality shuttlecock.

Section snippets

Related works

Feather detection can be divided into parameter and defect detection. The shuttlecock technique requires the implantation of only feather pieces with similar parameters. Thus, the parameters of these feather pieces must be measured. Particularly, the arching factor, camber, thickness, and width of the feather pieces are the main indicators that must be detected. At present, the feather piece parameters can be measured through artificial and machine vision technologies. Artificial vision

Image enhancement based on fuzzy set

Image enhancement techniques can highlight the interesting information of an image while attenuating other valuable data. The gray histogram of the feather piece (Fig. 5) shows that the gray value of the feather leaf has a large overlap with the gray value of the feather quill. This problem prevents the image segmentation method based on the image gray feature, such as the edge-based algorithms that use inter-region grayscale discontinuities and threshold-based algorithms that utilize grayscale

Feature modeling

For the grayscale image I(x,y), $F$ is used to represent the image features extracted from image I, as shown as follows: $F (x, y) = ϕ (I, x, y),$ where $ϕ$ represents the mapping of image attributes. The covariance feature matrix is used to describe the target to be classified. $ϕ$ is defined as follows: $ϕ (I, x, y) = {[x, y, I (x, y), |I_{x}|, |I_{y}|]}^{T}$ $ϕ (I, x, y) = {[x, y, I (x, y), |I_{x}|, |I_{y}|, arctan (|I_{x}| / |I_{y}|)]}^{T}$ $ϕ (I, x, y) = {[x, y, I (x, y), |I_{x}|, |I_{y}|, |I_{xx}|, |I_{yy}|]}^{T}$ $ϕ (I, x, y) = {[x, y, I (x, y), |I_{x}|, |I_{y}|, |I_{xx}|, |I_{yy}|, arctan (|I_{x}| / |I_{y}|)]}^{T}$ $ϕ (I, x, y) = {[x, y, I (x, y), |I_{x}|, |I_{y}|, |I_{xx}|, |I_{yy}|, arctan (|I_{x}| / |I_{y}|), \sqrt{{(I_{x})}^{2} + {(I_{y})}^{2}}]}^{T}$

In the expression, $I (x, y)$ is the

Experiment results and analysis

The defects of feather pieces are featured by color, geometry, and texture. Generally, relying on a single category of features can insufficiently distinguish the feather category accurately. Fig. 7 shows the partial sample images with different defects. In the subsequent recognition and classification, the defect region is only retained for these defect samples, and the gray value of the non-defect region is 0 during preprocessing.

Conclusions

Defect detection of feather pieces used in the production of shuttlecocks is a difficult process. The current method is performed manually. Machine learning is widely used in industry detection to improve production efficiency and ensure product quality. Hence, the feather image based on the fuzzy set is enhanced, and a detection method of feather defect combining Riemannian manifold and FFLD is proposed to achieve the effective defect classification of feather leaves. The conclusions in this

CRediT authorship contribution statement

Hongwei Yue: Writing - original draft, Methodology, Software. Hongtao Wang: Data curation, Conceptualization. Huazhou Chen: Writing - review & editing, Investigation. Ken Cai: Writing - review & editing, Supervision. Yingying Jin: Methodology.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgement

This work was supported in part by the China Postdoctoral Science Foundation under Grant 2018T110880, Grant 2017M620375, in part by the Natural Science Foundation of Guangdong Province under Grant 2018A030313882, Grant 2016A030313003, in part by the National Natural Science Foundation of China under Grant 81671788, Grant 61505037, in part by the Projects for International Scientific and Technological Cooperation under Grant 2018A05056084, in part by the Science Foundation for Young Teachers of

References (27)

C.S.H. Lin et al.
Aerodynamics of badminton shuttlecock: characterization of flow around a conical skirt with gaps, behind a hemispherical dome
J. Wind Eng. Ind. Aerodyn.
(2014)
Z. Chen et al.
A novel sparse representation model for pedestrian abnormal trajectory understanding
Expert Syst. Appl.
(2019)
K. Cai et al.
A framework combining window width-level adjustment and Gaussian filter-based multi-resolution for automatic whole heart segmentation
Neurocomputing
(2017)
P.T. Fletcher et al.
Riemannian geometry for the statistical analysis of diffusion tensor data
Signal Process.
(2007)
N. Mahmoudi et al.
Detecting credit card fraud by modified Fisher discriminant analysis
Expert Syst. Appl.
(2015)
F. Alam et al.
Aerodynamic properties of badminton shuttlecock
Int. J. Mech. Mater. Eng.
(2009)
S. Sturgess et al.
Design and implementation of a specific strength program for badminton
Strength Cond. J
(2008)
X. Ai et al.
The application of roundness error measurement in the detection of badminton appearance
J. Guangdong Univ. Technol
(2011)
H. Yue et al.
Kernel sparse representation on Riemannian manifold and its application to crease recognition
J. Chin. Mini-Micro Comp. Syst
(2015)
H. Liu et al.
Improved algorithm of feather image segmentation based on active contour model
J. Comp. Appl
(2011)

J. Chen et al.

Automatic defect detection of fasteners on the catenary support device using deep convolutional neural network

IEEE Trans. Instrum. Meas.

(2018)

D. Carrera et al.

Defect detection in SEM images of nanofibrous materials

IEEE Trans.Ind. Informat.

(2017)

K. Cai et al.

Tetralogy of Fallot cardiac function evaluation and intelligent diagnosis based on dual-source computed tomography cardiac images

Artif. Organs

(2016)

Cited by (58)

Modeling and estimation of water activity for the ionic-liquid-based aqueous ternary systems by smart paradigms
2024, Journal of the Taiwan Institute of Chemical Engineers
In this study, different artificial intelligence (AI) techniques are employed to determine the water activity of ionic liquid-based ternary systems as a function of pressure, temperature, and the type and concentration of the components. The proposed model has been selected among five artificial neural networks (ANNs), bagged trees (BT), decision trees (DT), gaussian process regression (GPR), adaptive neuro-fuzzy inference systems (ANFIS), and least square support vector machines (LS-SVM) by trial-and-error process and ranking analysis of statistical and computational indicators. The most accurate model is the cascade feed-forward neural network with fifteen hidden neurons which are equipped with logarithm and hyperbolic tangent sigmoid transfer functions. This model estimates 1829 empirical water activity for ionic liquid-based ternary systems with AARD=0.25 %, MSE=1.73 × 10⁻⁵, RMSE=4.16 × 10⁻³, and R²=0.9919.
A pixel-level deep segmentation network for automatic defect detection
2023, Expert Systems with Applications
Defect detection is a very important link for much manufacturing and processing applications which could be used for quality control and precise maintenance decision. However, faced with the weak-texture and low-contrast industrial environment, high-precision defect detection still faces a certain challenge due to diverse and complex of defects. Meanwhile, due to a minimal portion image pixels of defects, the pixel-level defect detection task is always against class-unbalance issue which also will affect the detection performance. Recently, with the strong automatic feature representation ability, deep learning has shown an excellent detection performance on defect identification and location. Nevertheless, it still has some demerits, such as insufficient processing of feature maps, lack of temporal modeling information, etc. To address these issues, on the basis of the encoder–decoder architecture, a pixel-level deep segmentation network is proposed for automatic defect detection to construct an end-to-end defect segmentation model. To realize effective feature representation, a residual attention network is proposed to construct the backbone network, which could also make the segmentation network better emphasize target regions. Meanwhile, to improve the network propagation ability of subtle context features, a bidirectional convolutional long short-term memory (ConvLSTM) block is introduced to optimize the skip connections to learn long-range spatial contexts. Besides, a weighted loss function is proposed for model training to address the class-unbalance issue. Combined with multiple public data sets, through qualitative and quantitative analysis, experimental results demonstrate that the proposed defect segmentation network achieves a better performance compared to other state-of-the-art segmentation methods.
An intelligent UAV path planning optimization method for monitoring the risk of unattended offshore oil platforms
2022, Process Safety and Environmental Protection
To ensure the safe operation of static and dynamic devices in a limited platform space and prevent security threats caused by accidents and external intruders, the unmanned aerial vehicle (UAV) is becoming an important tool to monitor and reduce the operational risk of unattended offshore oil platforms. However, UAVs may encounter location errors, adversely affecting patrol efficiency and accuracy. Therefore, the optimal UAV path planning is vital to ensure complete monitoring routes and reduce the risk presented by the marine environment, devices, and employees on unattended offshore platforms. This paper established a multi-objective mathematical model with the shortest flight path and minimum correction times for the intelligent UAV patrol of unattended offshore platforms. An intelligent algorithm with large-scale constraints for UAV path planning was proposed based on non-dominated sorting genetic algorithms. A flight path length of 8076.11 m and 25 correction times presented the optimal solution. The results indicated that the proposed algorithm could be used to plan an effective and accurate three-dimensional (3D) UAV flight path according to the size of the offshore oil platform. This is highly significant for the intelligent risk monitoring of unattended offshore platforms in practical engineering in the future.
An image-level weakly supervised segmentation method for No-service rail surface defect with size prior
2022, Mechanical Systems and Signal Processing
Citation Excerpt :
Machine vision inspection is an advanced non-destructive testing method with its intuitiveness and rapidity. It has been applied to detect surface defects in some fields such as shuttlecock [1], metallic component [2,3], metallic beam [17,18], steel strip [23,24,38] and pavement [39]. In recent years, many inspection methods based on computer vision have been proposed for rail surface defect detection because of its fast and intuition.
Surface defect segmentation of no-service rail is important for its quality assessment. Several deep learning methods have obtained good performances based on computer vision testing. However, these methods all require a large number of training samples with pixel-level labels. Marking pixel-level labels is a difficult task – time-consuming, labor-consuming, and professionals. To address the issue, a novel image-level weakly supervised segmentation formulation is proposed for no-service rail surface defects. These defects are decomposed into three sub-categories (strip-shaped, spot-shaped, block-shaped) according to the size prior information (area and shape). Then, a method is presented with a pooling combination module. The pooling combination module makes full use of the size attributes of the sub-category by utilizing different pooling functions for different sub-categories. Experimental results demonstrate that our method is effective and outperforms the state-of-the-art methods.
Index terms - Surface defect segmentation; no-service rail; image-level weakly supervised; size prior information; pooling combination module.
Enhanced Harris hawks optimization with multi-strategy for global optimization tasks
2021, Expert Systems with Applications
Harris Hawks Optimization (HHO) algorithm is a newly proposed meta-heuristic optimization algorithm that simulates the hunting process of the Harris hawks. It has the characteristics of fewer adjustment parameters and a strong optimization effect, resulting in strong competitiveness in similar optimization algorithms. However, HHO is prone to premature convergence and low convergence accuracy when dealing with specific complex optimization problems. Therefore, our work integrates two novel strategies into the standard HHO to gain enhanced exploration and exploitation capabilities. Specifically, our work firstly proposed an exploration strategy based on logarithmic spiral and opposition-based learning to improve the exploration ability of HHO. Secondly, the local search technique for Rosenbrock Method (RM) is modified to dynamically fuse into the standard HHO to enhance the algorithm’s local search capability and improve the convergence accuracy. The novel meta-heuristic algorithm proposed in this paper is called RLHHO. Finally, to validate the algorithm’s effectiveness, the proposed RLHHO algorithm is fully performance tested with eight other traditional meta-heuristic optimization algorithms on 23 benchmark functions and 30 IEEE CEC2014 test functions. Besides, another six advanced meta-heuristics algorithms are also compared in the 30 CEC’2014 test functions. The experimental results show that RLHHO performs significantly better than HHO as well as other traditional and advanced meta-heuristic algorithms in most of the test functions. To test the scalability of RLHHO in complex real-world problems, it was used to optimize the solution of three constrained real-world engineering problems, and the experimental results show that RLHHO’s powerful performance can be used as an effective tool for solving constrained engineering problems. Also, an effective hybrid model of kernel extreme learning machine is developed on the basis of RLHHO to cope with bankruptcy prediction problem. The experimental results show that this hybrid model is highly competitive with other mainstream classifiers regarding stability and prediction accuracy. The supplementary info and answers to possiblequeries will be publicly available at https://www.researchgate.net/profile/Chenyang_Li39/research.
Modelling of piping collapses and gully headcut landforms: Evaluating topographic variables from different types of DEM
2021, Geoscience Frontiers
The geomorphic studies are extremely dependent on the quality and spatial resolution of digital elevation model (DEM) data. The unique terrain characteristics of a particular landscape are derived from DEM, which are responsible for initiation and development of ephemeral gullies. As the topographic features of an area significantly influences on the erosive power of the water flow, it is an important task the extraction of terrain features from DEM to properly research gully erosion. Alongside, topography is highly correlated with other geo-environmental factors i.e. geology, climate, soil types, vegetation density and floristic composition, runoff generation, which ultimately influences on gully occurrences. Therefore, terrain morphometric attributes derived from DEM data are used in spatial prediction of gully erosion susceptibility (GES) mapping. In this study, remote sensing-Geographic information system (GIS) techniques coupled with machine learning (ML) methods has been used for GES mapping in the parts of Semnan province, Iran. Current research focuses on the comparison of predicted GES result by using three types of DEM i.e. Advanced Land Observation satellite (ALOS), ALOS World 3D-30 m (AW3D30) and Advanced Space borne Thermal Emission and Reflection Radiometer (ASTER) in different resolutions. For further progress of our research work, here we have used thirteen suitable geo-environmental gully erosion conditioning factors (GECFs) based on the multi-collinearity analysis. ML methods of conditional inference forests (Cforest), Cubist model and Elastic net model have been chosen for modelling GES accordingly. Variable’s importance of GECFs was measured through sensitivity analysis and result show that elevation is the most important factor for occurrences of gullies in the three aforementioned ML methods (Cforest = 21.4, Cubist = 19.65 and Elastic net = 17.08), followed by lithology and slope. Validation of the model’s result was performed through area under curve (AUC) and other statistical indices. The validation result of AUC has shown that Cforest is the most appropriate model for predicting the GES assessment in three different DEMs (AUC value of Cforest in ALOS DEM is 0.994, AW3D30 DEM is 0.989 and ASTER DEM is 0.982) used in this study, followed by elastic net and cubist model. The output result of GES maps will be used by decision-makers for sustainable development of degraded land in this study area.

View all citing articles on Scopus

¹: These authors contributed equally to this work and should be considered co-first authors.

View full text

Automatic detection of feather defects using Lie group and fuzzy Fisher criterion for shuttlecock production

Highlights

Abstract

Introduction

Section snippets

Related works

Image enhancement based on fuzzy set

Feature modeling

Experiment results and analysis

Conclusions

CRediT authorship contribution statement

Declaration of Competing Interest

Acknowledgement

J. Wind Eng. Ind. Aerodyn.

Expert Syst. Appl.

Neurocomputing

Signal Process.

Expert Syst. Appl.

Aerodynamic properties of badminton shuttlecock

Int. J. Mech. Mater. Eng.

Design and implementation of a specific strength program for badminton

Strength Cond. J

The application of roundness error measurement in the detection of badminton appearance

J. Guangdong Univ. Technol

Kernel sparse representation on Riemannian manifold and its application to crease recognition

J. Chin. Mini-Micro Comp. Syst

Improved algorithm of feather image segmentation based on active contour model