CNN performance dependence on linear image processing

2020 ◽  
Vol 2020 (10) ◽  
pp. 310-1-310-7
Author(s):  
Khalid Omer ◽  
Luca Caucci ◽  
Meredith Kupinski
Keyword(s):  
Image Processing ◽  
Texture Classification ◽  
Full Rank ◽  
Detection Performance ◽  
Ideal Observer ◽  
Training Data ◽  
Image Texture ◽  
Training Images ◽  
Analytic Expressions ◽  
Linear Compression

This work reports on convolutional neural network (CNN) performance on an image texture classification task as a function of linear image processing and number of training images. Detection performance of single and multi-layer CNNs (sCNN/mCNN) are compared to optimal observers. Performance is quantified by the area under the receiver operating characteristic (ROC) curve, also known as the AUC. For perfect detection AUC = 1.0 and AUC = 0.5 for guessing. The Ideal Observer (IO) maximizes AUC but is prohibitive in practice because it depends on high-dimensional image likelihoods. The IO performance is invariant to any fullrank, invertible linear image processing. This work demonstrates the existence of full-rank, invertible linear transforms that can degrade both sCNN and mCNN even in the limit of large quantities of training data. A subsequent invertible linear transform changes the images’ correlation structure again and can improve this AUC. Stationary textures sampled from zero mean and unequal covariance Gaussian distributions allow closed-form analytic expressions for the IO and optimal linear compression. Linear compression is a mitigation technique for high-dimension low sample size (HDLSS) applications. By definition, compression strictly decreases or maintains IO detection performance. For small quantities of training data, linear image compression prior to the sCNN architecture can increase AUC from 0.56 to 0.93. Results indicate an optimal compression ratio for CNN based on task difficulty, compression method, and number of training images.

scholarly journals Limitations of CNNs for Approximating the Ideal Observer Despite Quantity of Training Data or Depth of Network

2020 ◽  
Author(s):  
Khalid Omer ◽  
Luca Caucci ◽  
Meredith Kupinski
Keyword(s):  
Operating Characteristic ◽  
Full Rank ◽  
Ideal Observer ◽  
Training Data ◽  
Image Texture ◽  
Training Images ◽  
Network Layers ◽  
Linear Transform ◽  
Texture Detection ◽  
The Ideal

The performance of a convolutional neural network (CNN) on an image texture detection task as a function of linear image processing and the number of training images is investigated. Performance is quantified by the area under (AUC) the receiver operating characteristic (ROC) curve. The Ideal Observer (IO) maximizes AUC but depends on high-dimensional image likelihoods. In many cases, the CNN performance can approximate the IO performance. This work demonstrates counterexamples where a full-rank linear transform degrades the CNN performance below the IO in the limit of large quantities of training data and network layers. A subsequent linear transform changes the images’ correlation structure, improves the AUC, and again demonstrates the CNN dependence on linear processing. Compression strictly decreases or maintains the IO detection performance while compression can increase the CNN performance especially for small quantities of training data. Results indicate an optimal compression ratio for the CNN based on task difficulty, compression method, and number of training images.

scholarly journals IDENTIFIKASI CITRA BATIK DENGAN METODE CONVOLUTIONAL NEURAL NETWORK

BUANA ILMU ◽  
2021 ◽  
Vol 6 (1) ◽  
pp. 192-208
Author(s):  
Ayu Ratna Juwita ◽  
Tohirn Al Mudzakir ◽  
Adi Rizky Pratama ◽  
Purwani Husodo ◽  
Rahmat Sulaiman
Keyword(s):  
Neural Network ◽  
Image Processing ◽  
Convolutional Neural Network ◽  
Image Classification ◽  
Training Data ◽  
Classification Model ◽  
Gray Level ◽  
Training Images ◽  
Accuracy Level ◽  
Occurrence Matrix

Batik merupakan suatu kerjianan tangan yang memiliki nilai seni yang cukup tinggi dan juga salah satu bagian dari budaya indonessia. Untuk melestraikan budaya warisan batik dapat dikakukan dengan berbagai cara dengan pengenalan pola batik yang sangat beragam khususnya batik karawang. Penelitian ini membahas klasifikasi pola batik karawang menggunakan Convolutional Neural Network (CNN)  dengan ciri gray level Co-ocurrence Matrix. Proses awal yang akan dilakukan  yaitu preprocessing untuk mengubah citra warna menjadi grayscale, selanjutnya citra akan di segmentasikan sehingga memisahkan citra pola batik dengan background menggunakan metode otsu dan di ekstraksi menggunakan metode gray level co-ocurrence matrix untuk mendeteksi pola-pola batik. selanjutnya akan diklasifikasikan menggunakan metode Convolutional Neural Network (CNN) yang memberikan hasil klasifikasi citra batik. Dengan penerapan model klasifikasi citra batik Karawang ini memliki data training sebanyak 1094 citra latih dengan nilai akurasi 18,19% untuk citra latih,  citra dapat mengklasifikasikan dengan uji coba 344 citra batik, 45 citra batik Karawang, 299 citra batik luar Karawang mencapai 18,60% nilai tingkat akurasi, sedangkan hasil uji coba menggunakan citra batik karawang yang dapat dikenali dan diklasifikasikan mencapai nilai tingkat akurasi 73,33 %. Kata Kunci : Klasifikasi citra batik, CNN, GLCM, Otsu, Image Processing   Batik is a handicraft that has a high artistic value and also Batik is a part of Indonesian culture. To preserve the cultural heritage of batik it can be do in various ways with the introduction of many diverse batik patterns, especially karawang batik.. This study discusses the classification of Karawang batik patterns using Convolutional Neural Network (CNN) with gray level co-occurrence matrix characteristics. Initial process is preprocessing to convert the color image to grayscale, Then the image will be segmented. It can separated the image of the batik pattern from the background using the Otsu method and extracted using the gray level co-occurrence matrix method to detect batik patterns. Then, it will be classified using the Convolutional Neural Network (CNN) method which gives the results of batik image classification. With the application of this Karawang batik image classification model, it has training data of 1094 training images with an accuracy value of 18.19% for training images, images can be classified by testing 344 batik images, 45 Karawang batik images, 299 outer Karawang batik images reaching 18.60 % the value of the accuracy level, while the results of the trial using the image of batik karawang which can be recognized and classified reach an accuracy level of 73.33%. Keywords: Batik image classification, CNN, GLCM, Otsu, Image Processing

scholarly journals Rethinking the Random Cropping Data Augmentation Method Used in the Training of CNN-Based SAR Image Ship Detector

2020 ◽  
Vol 13 (1) ◽  
pp. 34
Author(s):  
Rong Yang ◽  
Robert Wang ◽  
Yunkai Deng ◽  
Xiaoxue Jia ◽  
Heng Zhang
Keyword(s):  
Neural Network ◽  
Data Augmentation ◽  
Back Propagation ◽  
Detection Performance ◽  
Training Data ◽  
Sar Image ◽  
Optical Images ◽  
The Neural Network ◽  
Effective Training ◽  
Standard Configuration

The random cropping data augmentation method is widely used to train convolutional neural network (CNN)-based target detectors to detect targets in optical images (e.g., COCO datasets). It can expand the scale of the dataset dozens of times while consuming only a small amount of calculations when training the neural network detector. In addition, random cropping can also greatly enhance the spatial robustness of the model, because it can make the same target appear in different positions of the sample image. Nowadays, random cropping and random flipping have become the standard configuration for those tasks with limited training data, which makes it natural to introduce them into the training of CNN-based synthetic aperture radar (SAR) image ship detectors. However, in this paper, we show that the introduction of traditional random cropping methods directly in the training of the CNN-based SAR image ship detector may generate a lot of noise in the gradient during back propagation, which hurts the detection performance. In order to eliminate the noise in the training gradient, a simple and effective training method based on feature map mask is proposed. Experiments prove that the proposed method can effectively eliminate the gradient noise introduced by random cropping and significantly improve the detection performance under a variety of evaluation indicators without increasing inference cost.

scholarly journals Implementasi Algoritma You Only Look Once (YOLO) untuk Deteksi Korban Bencana Alam

2021 ◽  
Vol 8 (4) ◽  
pp. 787
Author(s):  
Moechammad Sarosa ◽  
Nailul Muna
Keyword(s):  
Image Processing ◽  
Natural Disasters ◽  
Natural Disaster ◽  
Test Data ◽  
Detection System ◽  
Training Data ◽  
Test Results ◽  
Disaster Victim ◽  
Rapid Evacuation

<p class="Abstrak">Bencana alam merupakan suatu peristiwa yang dapat menyebabkan kerusakan dan menciptakan kekacuan. Bangunan yang runtuh dapat menyebabkan cidera dan kematian pada korban. Lokasi dan waktu kejadian bencana alam yang tidak dapat diprediksi oleh manusia berpotensi memakan korban yang tidak sedikit. Oleh karena itu, untuk mengurangi korban yang banyak, setelah kejadian bencana alam, pertama yang harus dilakukan yaitu menemukan dan menyelamatkan korban yang terjebak. Penanganan evakuasi yang cepat harus dilakukan tim SAR untuk membantu korban. Namun pada kenyataannya, tim SAR mengalami kendala selama proses evakuasi korban. Mulai dari sulitnya medan yang dijangkau hingga terbatasnya peralatan yang dibutuhkan. Pada penelitian ini sistem diimplementasikan untuk deteksi korban bencana alam yang bertujuan untuk membantu mengembangkan peralatan tim SAR untuk menemukan korban bencana alam yang berbasis pengolahan citra. Algoritma yang digunakan untuk mendeteksi ada atau tidaknya korban pada gambar adalah <em>You Only Look Once</em> (YOLO). Terdapat dua macam algoritma YOLO yang diimplementasikan pada sistem yaitu YOLOv3 dan YOLOv3 Tiny. Dari hasil pengujian yang telah dilakukan didapatkan <em>F1 Score</em> mencapai 95.3% saat menggunakan YOLOv3 dengan menggunakan 100 data latih dan 100 data uji.</p><p class="Abstrak"> </p><p class="Abstrak"><strong><em>Abstract</em></strong></p><p class="Abstrak"> </p><p class="Abstract"><em>Natural disasters are events that can cause damage and create havoc. Buildings that collapse and can cause injury and death to victims. Humans can not predict the location and timing of natural disasters. After the natural disaster, the first thing to do is find and save trapped victims. The handling of rapid evacuation must be done by the SAR team to help victims to reduce the amount of loss due to natural disasters. But in reality, the process of evacuating victims of natural disasters is still a lot of obstacles experienced by the SAR team. It was starting from the difficulty of the terrain that is reached to the limited equipment needed. In this study, a natural disaster victim detection system was designed using image processing that aims to help find victims in difficult or vulnerable locations when directly reached by humans. In this study, a detection system for victims of natural disasters was implemented which aims to help develop equipment for the SAR team to find victims of natural disasters based on image processing. The algorithm used is You Only Look Once (YOLO). In this study, two types of YOLO algorithms were compared, namely YOLOv3 and YOLOv3 Tiny. From the test results that have been obtained, the F1 Score reaches 95.3% when using YOLOv3 with 100 training data and 100 test data.</em></p>

scholarly journals Identifikasi Kematangan Cabai Menggunakan Operasi Morfologi (Opening dan Closing) dan Metode Backpropagation

SISTEMASI ◽  
2021 ◽  
Vol 10 (1) ◽  
pp. 96
Author(s):  
Khairullah Khairullah ◽  
Erwin Dwika Putra
Keyword(s):  
Image Processing ◽  
Digital Image Processing ◽  
Confusion Matrix ◽  
Training Data ◽  
Morphological Operations ◽  
Morphological Operation ◽  
Backpropagation Method ◽  
Grouping Method ◽  
Opening And Closing

AbstrakIdentifikasi kualitas buah cabai biasanya masih menggunakan cara visual secara langsung atau sortir secara manual oleh petani, dengan menggunakan sistem ini sering kali terjadi beberapa kesalahan setiap melakukan sortir yang disebabkan oleh petani yang melakukan sortir merasa terlalu lelah. Dengan menggunakan komputasi pengolahan citra digital, untuk melakukan identifikasi pengelompokan buah cabai yang matang dan mentah dapat membantu para petani, Teknik pengelompokan ini akan menggunakan metode pengelompokan berdasarkan warna. Metode pengelompokan tersebut sebelumnya akan dilakukan operasi morfologi pada citra yang telah diambil. Pendekatan operasi morfologi pada penelitian ini adalah Opening and Closing, pada operasi morfologi akan menghilangkan noise dan menebalkan objek dari inputan gambar. Metode Bacpropagatioan akan mengolah data latih sebanyak 10 data latih mendapatkan 6 iterasi perhitungan dan setelah diuji menggunakan data uji hasil yang didapatkan yaitu tingkat pengenalan rata-rat mendapatkan perhitungan sebanyak 7 iterasi metode Bacpropagation. Hasil dari penelitian ini juga dihitung menggunakan Confusion Matrix dimana nilai Precision 90%, Recall 74%, dan Accuracy 70%, maka dapat disimpulkan bahwa Operasi Morfologi dan Metode Backpropagation dapat digunakan untuk mengidentifikasi objek cabai.Kata Kunci: backpropagation, morfologi, identifikasi, opening and closing  AbstractIdentification of the quality of chili fruit is usually still using a visual way directly or sorting manually by farmers, using this system often occurs several errors, every sorting caused by farmers who do the sorting feel too tired. By using digital image processing computing, to identify the grouping of ripe and raw chili fruits can help farmers, this grouping technique will use a method of grouping based on color. The grouping method will previously perform morphological surgery on the image that has been taken. The morphological operation approach in this study is Opening and Closing, in morphological operations will eliminate noise and thicken objects from image input. Bacpropagatioan method will process training data as much as 10 training data get 6 iterations of calculations and after being tested using the test data obtained results that is the level of introduction of the average rat get a calculation of 7 iterations bacpropagation method. The results of this study were also calculated using Confusion Matrix where precision values of 90%, Recall 74%, and Accuracy 70%, it can be concluded that Morphological Operations and Backpropagation Method can be used to identify chili objects.Keywords: backpropagation, morfologi, identification, opening and closing

Implementation of Backpropagation for Ulap-ulap Pattern Recognition

2021 ◽  
Vol 9 (3) ◽  
pp. 405
Author(s):  
Ni Luh Yulia Alami Dewi ◽  
I Wayan Santiyasa
Keyword(s):  
Image Processing ◽  
Pattern Recognition ◽  
Data Collection ◽  
Test Data ◽  
Training Data ◽  
Image Pattern

Ulap-ulap is one of the symbols used to indicate that a building has been carried out Mlaspas ceremony. Mlaspas is one of the ceremonies performed to purify and clean a building. Ulap-ulap itself consists of various types depending on the building where it is placed, for example the ulap-ulap placed on the Pelinggih building will be different from the ulap-ulap placed on the Bale building. So that the pattern contained in each type of Ulap-ulap is different. The purpose of this research is to be able to do pattern recognition on Ulap-ulap images. The method used in this study is Backpropagation, and for its implementation, the MATLAB 7.5.0 (R2007b) application is used. This study used 18 images of Ulap-ulap, including 15 training data and 6 test data. The stages of the process carried out are for Ulap-ulap pattern recognition, the first is data collection, then image processing, and finally the pattern recognition. Recognition of the Ulap-ulap image pattern with Backpropagation, resulted in an accuracy of 83.333%.

scholarly journals Improving Regional and Teleseismic Detection for Single-Trace Waveforms Using a Deep Temporal Convolutional Neural Network Trained with an Array-Beam Catalog

Sensors ◽  
2019 ◽  
Vol 19 (3) ◽  
pp. 597 ◽  
Author(s):  
Joshua Dickey ◽  
Brett Borghetti ◽  
William Junek
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Detection Performance ◽  
Training Data ◽  
Characteristic Functions ◽  
Type I ◽  
Test Set ◽  
Arrival Times ◽  
Treaty Monitoring ◽  
Single Trace

The detection of seismic events at regional and teleseismic distances is critical to Nuclear Treaty Monitoring. Traditionally, detecting regional and teleseismic events has required the use of an expensive multi-instrument seismic array; however in this work, we present DeepPick, a novel seismic detection algorithm capable of array-like detection performance from a single-trace. We achieve this performance through three novel steps: First, a high-fidelity dataset is constructed by pairing array-beam catalog arrival-times with single-trace waveforms from the reference instrument of the array. Second, an idealized characteristic function is created, with exponential peaks aligned to the cataloged arrival times. Third, a deep temporal convolutional neural network is employed to learn the complex non-linear filters required to transform the single-trace waveforms into corresponding idealized characteristic functions. The training data consists of all arrivals in the International Seismological Centre Database for seven seismic arrays over a five year window from 1 January 2010 to 1 January 2015, yielding a total training set of 608,362 detections. The test set consists of the same seven arrays over a one year window from 1 January 2015 to 1 January 2016. We report our results by training the algorithm on six of the arrays and testing it on the seventh, so as to demonstrate the generalization and transportability of the technique to new stations. Detection performance against this test set is outstanding, yielding significant improvements in recall over existing techniques. Fixing a type-I error rate of 0.001, the algorithm achieves an overall recall (true positive rate) of 56% against the 141,095 array-beam arrivals in the test set, yielding 78,802 correct detections. This is more than twice the 37,572 detections made by an STA/LTA detector over the same period, and represents a 35% improvement over the 58,515 detections made by a state-of-the-art kurtosis-based detector. Furthermore, DeepPick provides at least a 4 dB improvement in detector sensitivity across the board, and is more computationally efficient, with run-times an order of magnitude faster than either of the other techniques tested. These results demonstrate the potential of our algorithm to significantly enhance the effectiveness of the global treaty monitoring network.

Inverse Kinematics Modelling and Simulation for Upper Case Writing Robot Control Using ANFIS

2016 ◽  
Vol 836 ◽  
pp. 37-41 ◽  
Author(s):  
Adlina Taufik Syamlan ◽  
Bambang Pramujati ◽  
Hendro Nurhadi
Keyword(s):  
Image Processing ◽  
Character Recognition ◽  
Inverse Kinematics ◽  
Robot Control ◽  
Industrial Revolution ◽  
Training Data ◽  
Robotic Arm ◽  
End Effector ◽  
Precision And Accuracy ◽  
Case Writing

Robotics has lots of use in the industrial world and has lots of development since the industrial revolution, due to its qualities of high precision and accuracy. This paper is designed to display the qualities in a form of a writing robot. The aim of this study is to construct the system based on data gathered and to develop the control system based on the model. There are four aspects studied for this project, namely image processing, character recognition, image properties extraction and inverse kinematics. This paper served as discussion in modelling the robotic arm used for writing robot and generating theta for end effector position. Training data are generated through meshgrid, which is the fed through anfis.

Fast Face Recognition on GPUs

2014 ◽  
Vol 905 ◽  
pp. 543-547
Author(s):  
Yi Lei ◽  
Xiao Ya Fan ◽  
Meng Zhang
Keyword(s):  
Image Processing ◽  
Principal Component Analysis ◽  
Pattern Recognition ◽  
Face Recognition ◽  
Processing Speed ◽  
Principal Component ◽  
Great Accuracy ◽  
Analysis Algorithm ◽  
Training Images ◽  
Principal Component Analysis Algorithm

Face recognition is popular in the field of pattern recognition and image processing. However, traditional recognition technologies spend too long there are a lot of images to be recognized or trained for great accuracy in the recognition. Parallel computing is an effective way to improve the processing speed. With the improvement of GPU performance, its widely applied in computing-concentrated data operations. This paper presents a study of performance speedup achieved by applying GPU for face recognition based on PCA (Principal Component Analysis) algorithm. We successfully accelerated the testing phase by 6868-folds compared to a sequential C implementation when it has 100 test images and 2400 training images.

Sign in / Sign up

Export Citation Format

Share Document