Accelerating nearest neighbor partitioning neural network classifier based on CUDA

Wang, Lin; Zhu, Xuehui; Yang, Bo; Guo, Jifeng; Liu, Shuangrong; Li, Meihui; Zhu, Jian; Abraham, Ajith

doi:10.1016/j.engappai.2017.10.023

Cited by 12 publications

(5 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…7. Third, rather than evaluating the photoacoustic spectra with an extensive look-up table (i.e., spectral atlas), the classification time can be reduced by implementing an artificial neural network (ANN) to learn and match features of the acoustic spectra (i.e., bypassing PCA feature extraction), which has been successfully implemented in previous studies (Jain and Mao, 1991;Wang et al, 2018).…”

Section: Discussionmentioning

confidence: 99%

Acoustic Frequency-Based Approach for Identification of Photoacoustic Surgical Biomarkers

2021

View full text Add to dashboard Cite

Spectral unmixing techniques for photoacoustic images are often used to isolate signal origins (e.g., blood, contrast agents, lipids). However, these techniques often require many (e.g., 12–59) wavelength transmissions for optimal performance to exploit the optical properties of different biological chromophores. Analysis of the acoustic frequency response of photoacoustic signals has the potential to provide additional discrimination of photoacoustic signals from different materials, with the added benefit of potentially requiring only a few optical wavelength emissions. This study presents our initial results testing this hypothesis in a phantom experiment, given the task of differentiating photoacoustic signals from deoxygenated hemoglobin (Hb) and methylene blue (MB). Coherence-based beamforming, principal component analysis, and nearest neighbor classification were employed to determine ground-truth labels, perform feature extraction, and classify image contents, respectively. The mean ± one standard deviation of classification accuracy was increased from 0.65 ± 0.16 to 0.88 ± 0.17 when increasing the number of wavelength emissions from one to two, respectively. When using an optimal laser wavelength pair of 710–870 nm, the sensitivity and specificity of detecting MB over Hb were 1.00 and 1.00, respectively. Results are highly promising for the differentiation of photoacoustic-sensitive materials with comparable performance to that achieved with more conventional multispectral laser wavelength approaches.

show abstract

Section: Discussionmentioning

confidence: 99%

Acoustic Frequency-Based Approach for Identification of Photoacoustic Surgical Biomarkers

2021

View full text Add to dashboard Cite

show abstract

“…Authors in [ 50 ] recommended a parallel nearest neighbor partitioning (NNP) method to accelerate NNP. In their method, blocks and threads are used to evaluate potential neural networks and to perform parallel subtasks.…”

Section: Review Of Some Related Workmentioning

confidence: 99%

Performance evaluation of spatial fuzzy C-means clustering algorithm on GPU for image segmentation

Ali

Abbassi

Bouattane

2022

Multimed Tools Appl

View full text Add to dashboard Cite

Image processing by segmentation technique is an important phase in medical imaging such as MRI. Its objective is to analyze the different tissues in human body. In research area, Fuzzy set is one of the most successful techniques that guarantees a robust classification. Spatial FCM (SFCM); one of the fuzzy c-means variants; considers spatial information to deal with the noisy images. To reduce this iterative algorithm’s execution time, a hard SIMD architecture has been planted named the Graphical Processing Unit (GPU). In this work, a great contribution has been done to diagnose, confront and implement three different parallel implementations on GPU. A parallel implementations’ extensive study of SFCM entitled PSFCM using 3 × 3 window is presented, and the experiments illustrate a significant decrease in terms of running time of this algorithm known by its high complexity. The experimental results indicate that the parallel version’s execution time is about 9.46 times faster than the sequential implementation on image segmentation. This gain in terms of speed-up is achieved on the Nvidia GeForce GT 740 m GPU.

show abstract

“…The performance of the proposed method, image moment anomaly detection based on Gaussian distribution model (IMA-GM), is evaluated by comparing with three common supervised classification algorithms, logistic regression (LR) [33], support vector machine (SVM) [34] and neural network (NN) [35]. In particular, LR is based on L 2 regularization and its penalty coefficient is set to 10 −5 ; the kernel function of SVM is a radial basis function and its kernel factor is 10; the NN has three layers and the node number of input, hidden and output layers are 32, 64 and 32 respectively, and the Sigmoid activation function is used.…”

Section: B Experiments 2: Comparative Test Using Different Defect Detmentioning

confidence: 99%

Automated Product Boundary Defect Detection Based on Image Moment Feature Anomaly

et al. 2019

View full text Add to dashboard Cite

Electric distribution cabinets are critical components in the power distribution pipeline. Surface defect detection plays an important role in the production process. It not only guarantees product quality but also affects the brand reputation. In particular, the boundaries of metallic cabinets are more vulnerable to be damaged than other surface areas. Thus, boundary defect detection is a bottleneck problem that needs to be solved. To deal with this issue, a method based on image moment feature anomaly is developed to detect the defects on cabinet surfaces. The boundary edges from an image of the produced cabinet are first extracted using a hybrid of edge detection and boundary skeleton extraction. Then, the boundary areas are divided into small and identical size image blocks. A Gaussian distribution model of normal image blocks without defects is established. Finally, the anomaly features of image blocks with defects are extracted to identify the defect image blocks based on the Gaussian distribution model and a segmentation threshold. Two experiments are carried out. One is to determine the optimal illumination intensity for image acquisition and the optimal threshold of defect detection. The other is to evaluate the performance of the defect detection method. This developed approach can be applied in the non-destructive and effective inspection of electric distribution cabinets and provides a feasible solution for metallic product quality assurance. INDEX TERMS Boundary defect detection, electric distribution cabinets, image moment, anomaly detection.

show abstract

Accelerating nearest neighbor partitioning neural network classifier based on CUDA

Cited by 12 publications

References 41 publications

Acoustic Frequency-Based Approach for Identification of Photoacoustic Surgical Biomarkers

Acoustic Frequency-Based Approach for Identification of Photoacoustic Surgical Biomarkers

Performance evaluation of spatial fuzzy C-means clustering algorithm on GPU for image segmentation

Automated Product Boundary Defect Detection Based on Image Moment Feature Anomaly

Contact Info

Product

Resources

About