A Deep Learning Framework for Multi-Class Objectrecognition Using Feature Fusion and Neigborhoodcomponent Analysis DOI
Anamika Dhillon, Gyanendra K. Verma

SSRN Electronic Journal, Journal Year: 2021, Volume and Issue: unknown

Published: Jan. 1, 2021

Object recognition is a computer vision technique for identifying objects inan image. Deep neural networks have demonstrated remarkable recognitionresults on the basis of features extracted from single image object. In this paper, we present feature fusion-based deep learning method forclassifying and recognizing multi-class objects. Specifically, first adopttwo Convolutional Neural Network models: DenseNet 201 ResNet 101, forfeature extraction. Then, to acquire more compact presentation featuresand reduce dimensions, utilized Neighborhood Component Analysis(NCA). Furthermore, fusion performed in hierarchical manner byapplying concatenation operation. Finally, classify multiple objectsin an by using Support Vector Machine (SVM) classifier. We demonstrate effectiveness our methodology two benchmark datasets; MSCOCO wild animal camera trap dataset. The experimental results showthat proposed framework achieved accuracy 98.1% 97.5% datasets respectively. Results showed thatour effectively improved performance favorably bothrobustness accuracy. fair comparison with existing techniques reported literature also provided

Language: Английский

A statistical categorization-based curriculum learning approach for multi-task classification of images DOI Creative Commons
Ozan Veranyurt, C. Okan Sakar

Applied Intelligence, Journal Year: 2025, Volume and Issue: 55(6)

Published: Feb. 28, 2025

Language: Английский

Citations

0
Recognition of Defective Carrots Based on Deep Learning and Transfer Learning DOI
Weijun Xie, Shuo Wei, Zhaohui Zheng

et al.

Food and Bioprocess Technology, Journal Year: 2021, Volume and Issue: 14(7), P. 1361 - 1374

Published: April 24, 2021

Language: Английский

Citations

30
Production of Cellulose Nanocrystals Suspension with High Yields from Water Hyacinth DOI Creative Commons
Pruttipong Pantamanatsopa, Warunee Ariyawiriyanan, Sanong Ekgasit

et al.

Journal of Natural Fibers, Journal Year: 2022, Volume and Issue: 20(1)

Published: Oct. 18, 2022

Acid hydrolysis is commonly used to extract cellulose nanocrystals (CNC) from natural fibers. This research thus investigates the effect of temperatures and time on yields CNC water hyacinth. The aim determine optimal acid condition for hyacinth fiber that effectively enhances yield. were varied between 50 60°C 15, 30, 60, 120 min. Prior hydrolysis, raw was treated with alkaline bleached by hydrogen peroxide. results showed 30 min, crystallinity index 80%, crystalline size 3.91 nm, yield 71.5%. transmission electron microscopy morphology whisker shape, diameter length 10 nm 200–500 nm. also indicated that, unlike time, temperature had a negligible index. Besides, under condition, stable aqueous suspension zeta potential −43.21 mV, indicating high physical colloidal stability.

Language: Английский

Citations

10
The use of machine learning for the determination of a type/model of firearms by the characteristics on cartridge cases DOI
Pavel Giverts, K. O. Sorokina, Mark Barash

et al.

Forensic Science International, Journal Year: 2024, Volume and Issue: 358, P. 112021 - 112021

Published: April 8, 2024

Language: Английский

Citations

1
Examination of the possibility to use Siamese networks for the comparison of firing pin marks DOI
Pavel Giverts, K. O. Sorokina, В. А. Федоренко

et al.

Journal of Forensic Sciences, Journal Year: 2022, Volume and Issue: 67(6), P. 2416 - 2424

Published: Sept. 23, 2022

One of the most discussed issues in forensic firearms identification is subjectivity conclusions. The main part examiners' work to make a microscopic comparison marks on cartridge cases and bullets. In this process, examiners have decide if quantity quality observed characteristics are sufficient for identification. This decision based personal experience an examiner, so with different backgrounds can come conclusions, fact presents problem. Besides, calculation error rate type examination debatable issue. Different mathematical statistical models were proposed, computer-based algorithms developed order avoid determine rates. article investigates possibility use methods machine learning firing pin impressions cases. research, Siamese network model, which included two similar Convolutional Neural Networks, was prepared trained. For training validation database prepared. images discharged from 300 that came regular casework clone used data augmentation. model trained examined using database. metrics, such as accuracy, sensitivity, specificity calculated. results research show building objective system known rate.

Language: Английский

Citations

6
Identification of bullets fired from air guns using machine and deep learning methods DOI Creative Commons
Muthu Rama Krishnan Mookiah, Roberto Puch‐Solis,

Niamh Nic Daéid

et al.

Forensic Science International, Journal Year: 2023, Volume and Issue: 349, P. 111734 - 111734

Published: May 19, 2023

Ballistics (the linkage of bullets and cartridge cases to weapons) is a common type evidence encountered in criminal around the world. The interest lies determining whether two were fired using same firearm. This paper proposes an automated method classify from surface topography Land Engraved Area (LEA) images pellets machine deep learning methods. curvature was removed loess fit features extracted Empirical Mode Decomposition (EMD) followed by various entropy measures. informative identified minimum Redundancy Maximum Relevance (mRMR), finally classification performed Support Vector Machines (SVM), Decision Tree (DT) Random Forest (RF) classifiers. results revealed good predictive performance. In addition, model DenseNet121 used LEA images. provided higher performance than SVM, DT RF Moreover, Grad-CAM technique visualise discriminative regions These suggest that proposed can be expedite projectiles firearms assist ballistic examinations. this work, compared air both rifles high velocity pistol. Air guns collect data because they more accessible other could as proxy, delivering comparable LEAs. methods developed here proof-of-concept are easily expandable bullet case identification any weapon.

Language: Английский

Citations

2
Scaled Dilation of DropBlock Optimization in Convolutional Neural Network for Fungus Classification DOI Open Access
Anuruk Prommakhot, Jakkree Srinonchat

Computers, materials & continua/Computers, materials & continua (Print), Journal Year: 2022, Volume and Issue: 72(2), P. 3313 - 3329

Published: Jan. 1, 2022

Image classification always has open challenges for computer vision research. Nowadays, deep learning promoted the development of this field, especially in Convolutional Neural Networks (CNNs). This article proposes efficiently scaled dilation DropBlock optimization CNNs fungus classification, which there are five species experiment. The proposed technique adjusts convolution size at 35, 45, and 60 with max-polling 2 × 2. models also designed 12 different BlockSizes KeepProp. techniques provide maximum accuracy 98.30% training set. Moreover, three accurate models, called Precision, Recall, F1-score, employed to measure testing experiment results expose that achieve classify an excellent compared previous techniques. Furthermore, can reduce structure layer, directly affecting resource time computation.

Language: Английский

Citations

3
Digital image analysis of gunshot residue dimensional dispersion by computer vision method DOI
İlker Kara,

Erşan Tahıllıoğlu

Microscopy Research and Technique, Journal Year: 2021, Volume and Issue: 85(3), P. 971 - 979

Published: Oct. 15, 2021

Detection and identification of gunshot residues (GSR) have been used as base evidence in elucidating forensic cases. GSR particles consist burnt partially unburned material contaminate the hands, face, hair, clothes shooter when coming out gun. Nowadays, samples are collected from hands suspect analyzed routinely laboratories by scanning electron microscope/energy dispersive spectroscopy (SEM/EDS) method. comprised a morphological specific structure (generally spherical diameter between 0 100 μm [occasionally even larger]). In addition, present studies field claimed that during formation formed under equilibrium surface distribution unrelated to dimensional classification. Our contribution this study is two-folded. First, offers new approach identify images computer vision gathered SEM/EDS method hand shooter. Second, it presents open access image data set GSR. During study, consisting 22,408 three different types MKEK (Mechanical Chemical Industries Corporation) brand ammunition has used. It seen results successful classification

Language: Английский

Citations

4
Development of an ai-enabled video capturing device for bullet trajectory analysis and ballistic research DOI Creative Commons

Shashanka Handique,

Sweta Saha,

R. Suresh

et al.

ITEGAM- Journal of Engineering and Technology for Industrial Applications (ITEGAM-JETIA), Journal Year: 2024, Volume and Issue: 10(47)

Published: Jan. 1, 2024

A ballistic experts' discipline is the ability to compare characteristic marks found on surface of different fired bullets determine whether they were from same gun. These tool become a "ballistic fingerprint" that examiners can use identify specific characteristics firearm discharged bullet. One such mark striation left bullet, identical scratch marks. Manually done, comparison microscope used in this process, where testing bullet rotated until well-defined land or groove comes into view. The sample then search matching region. But process opinions are given through only manual experimental and not an automated system. proposed solution was develop cost-effective system captures video one go. Also, focus lighting arrangement independent environment, so device be efficiently any environment.

Language: Английский

Citations

0
Exploring lightweight convolution neural networks for segmenting striation marks from firearm bullet images DOI

Genevieve Chyrmang,

Barun Barua,

Kangkana Bora

et al.

Forensic Imaging, Journal Year: 2024, Volume and Issue: 39, P. 200611 - 200611

Published: Nov. 13, 2024

Language: Английский

Citations

0