Papers by Sanparith Marukatat
Cornell University - arXiv, Jun 15, 2022
Fast, accurate and affordable rice disease detection method is required to assist rice farmers ta... more Fast, accurate and affordable rice disease detection method is required to assist rice farmers tackling equipment and expertise shortages problems. In this paper, we focused on the solution using computer vision technique to detect rice diseases from rice field photograph images. Dealing with images took in real-usage situation by general farmers is quite challenging 1
Exhaled volatile organic compounds for cholangiocarcinoma diagnosis
Liver Research
Combining Technical Indicators and Deep Learning by using LSTM Stock Price Predictor
2021 18th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON)
A stock market price prediction is one of the most challenging tasks. There are many effected fac... more A stock market price prediction is one of the most challenging tasks. There are many effected factors to make the stock market price go up or down such as dynamic environment, economy, politics. Thus, it is very difficult to predict a future price. Although, there are many technical indicators available for traders and investors at the present, the indicators might not be able to capture market behaviors. Nowadays, machine learning has more influence on the financial market prediction and trading strategies. This work introduced a stock market prediction by using the LSTM network model to predict future prices based on historical data. The model was trained in SET100 index from 2016 to 2019. Then, the model was tested through our trading operations on all stocks in SET100 in 2019 and 2020 to evaluate the profitable model. The experimental results show that the model is promising. The model in both long-only strategy and long-short strategy can achieve higher return than buy and hold, MACD, BB, RSI strategies both 2019 and 2020.
Text generation by probabilistic suffix tree language model
2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP), 2021
During last decade, language modeling has been dominated by neural structures; RNN, LSTM or Trans... more During last decade, language modeling has been dominated by neural structures; RNN, LSTM or Transformer. These neural language models provide excellent performance to the detriment of very high computational cost. This work investigates the use of probabilistic language model that requires much less computational cost. In particular, we are interested in variable-order Markov model that can be efficiently implemented on a probabilistic suffix tree (PST) structure. The PST construction is cheap and can be easily scaled to very large dataset. Experimental results show that this model can be used to generated realistic sentences.
MultiMedia Modeling, 2018
In this paper, we present the Sloth Search System (SSS) for large scale video browsing. Our key c... more In this paper, we present the Sloth Search System (SSS) for large scale video browsing. Our key concept is to apply object recognition and scene classification to generate keyword tags from video images. This indexing process is performed only on selected frames for faster processing. The keyword tags are used to retrieve videos from a text-based query. Additional feature signatures are also used to extract spatial and color information. These proposed signatures are stored as binary codes for a compact representation and for fast search. Such a representation allows users to search by drawing a sketch or a bounding box of a specific object.
Image-Based Silkworm Egg Classification and Counting Using Counting Neural Network
Silkworm egg classification and counting are essential tasks in the silkworm industry for promoti... more Silkworm egg classification and counting are essential tasks in the silkworm industry for promotion and conservation of the silkworm gene. Normally, the egg counting process is done by human or estimated from the average weight of an egg. However, these methods have been proven to be both time-consuming and inaccurate. Therefore, in this work, we develop a silkworm counting system that can count eggs laid on the disease-free laying (DFL) sheet image. The system can count eggs in all classes that are in the fresh, all-blue, and shell period. The result shows that the system yields approximately 80 to 88% counting rate in fresh and shell period. Whereas in the all-blue period, the system can produce about 60 to 78% counting rate because of the condition of the type of DFL sheet and the similar characteristic of all-blue in the early stage and unfertilized eggs.
Text detection and recognition on traffic panel in roadside imagery
2017 8th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES), 2017
Text recognition has revolutionized the world of image processing and intelligent transportation ... more Text recognition has revolutionized the world of image processing and intelligent transportation system (ITS). It opened several possibilities to traditional ITS concept. Advancement in text recognition has made it possible to implement text recognition in ITS. Traffic panel text recognition, a real time application is considered as a key addition to the revolution in modern ITS. This research aims at developing real time application for traffic panel text recognition for English and Thai. Text recognition in this paper is based on Support vector Machine (SVM), KNN and maximally stable extremal regions (MSER). Traffic panel are extracted based on visual appearance. Based on traffic panel background, color mask is applied to remove non text candidates. Multi-level MSER is used for segmentation. Raw pixel value of segmented character is used as feature vector. This system is trained for English and Thai language and then tested on roadside images of traffic panels. The results of different kernels are compared with each other to select best possible kernel for SVM. SVM results are then compared with KNN to find best classifier for this problem. The result shows that proposed system is robust and performs well in challenging environment.
On the Use of Attention Map for Land Cover Mapping
2020 17th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON), 2020
The use of machine learning technology with remote sensing image analysis, especially for the lan... more The use of machine learning technology with remote sensing image analysis, especially for the land cover mapping requires experts and huge resources because every pixel in the training set must be labeled. This task is time-consuming and tedious. Therefore, a better strategy is to only identify what classes are present in an image without specifying where they are. In this way, a large number of remote sensing images can be labeled quickly. To achieve this goal, we employed the attention layer to create the attention map. The attention map is then further segmented to produce the final l and c over m ap where every pixel in an image will be labeled. We have tested the performance of our proposed algorithm with UC Merced Dataset and achieved 79.7 % in identifying the presence of land cover classes and 71.2 % accuracy in the labeling of all pixels
A Simulated-Data Adaptation Technique
This paper proposes an efficient acoustic model adaptation method based on the use of simulated-d... more This paper proposes an efficient acoustic model adaptation method based on the use of simulated-data in maximum likelihood linear regression (MLLR) adaptation for robust speech recognition. Online MLLR adaptation is an unsupervised process which requires an input speech with phone labels transcribed automatically. Instead of using only the input signal in adaptation, our proposed simulated data method increases the size of adaptation data by adding noise portions extracted from the input speech to a set of pre-recorded clean speech, whose correct transcriptions are known. Various configurations of the proposed method are explored. Evaluations are performed with both additive and real noisy speech. The experimental results show that the proposed system achieves higher recognition rate than the system using only the input speech in adaptation and the system using a multi-conditioned acoustic model.
2007 International Symposium on Communications and Information Technologies, 2007
This paper proposes the use of tree-structured model selection and simulated-data in maximum like... more This paper proposes the use of tree-structured model selection and simulated-data in maximum likelihood linear regression (MLLR) adaptation for environment and speaker robust speech recognition. The objective of this work is to solve major problems in robust speech recognition system, namely unknown speaker and unknown environmental noise. The proposed solution is composed of two components. The first one is based on a tree-structured model for selecting a speakerdependent model that best matches to the input speech. The second component uses simulated-data to adapt the selected acoustic model to fit with the unknown noise. The proposed technique can thus alleviate both problems simultaneously. Experimental results show that the proposed system achieves a higher recognition rate than the system using only the input speech in adaptation and the system using a multi-conditioned acoustic model.
Attention Mechanism for Land Cover Mapping with Image-Level Labels
2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, 2021
In this paper, we address the problem of land cover mapping where only the image labels are avail... more In this paper, we address the problem of land cover mapping where only the image labels are available in a training set. Here, a training sample is a pair of a remote sensing image and a class indicator of whether a land cover class is present or absent on the entire image. It is more economical to only identify what classes are present in an image than to label every pixel in an image. In this way, a large number of remote sensing images can be labeled quickly. To achieve this goal, we employed the attention layer to create the attention map. The attention map is then further segmented to produce the final land cover map where every pixel in an image will be labeled. We have tested the performance of our proposed algorithm with the UC Merced Dataset and achieved more than 85% overall accuracy with only 0.4% labels in a training set when comparing with the UNET [1].
Image enhancement by pixels sorting
2018 15th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON)
Image enhancement is one of the most fundamental processing steps in many imaging systems. Histog... more Image enhancement is one of the most fundamental processing steps in many imaging systems. Histogram equalization is one of the widely used enhancement technique. It is based on the cumulative distribution function mapping. Unfortunately, due to the discrete nature of the histogram, the mapping procedure cannot guarantee to produce correctly equalized distribution. This paper proposes a simple procedure called the “pixels sorting” to fill in this gap. The proposed procedure can also enhance local detail of image having uneven illumination. It could be used with other brightness preserving enhancement techniques as well.
International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management, 2012
For centuries, in the North of Thailand, many books of Lanna Dharma characters had been printed. ... more For centuries, in the North of Thailand, many books of Lanna Dharma characters had been printed. These books are the important sources of the knowledge of ancient Lanna wisdom. At present, the books are found old and damaged. Most of characters are rough and not clear according to its early printing technology at that time. Moreover, some sets of characters are relatively very similar which cause the difficulty to recognize them. This paper proposes a Lanna Dharma printed character recognition technique using k-Nearest Neighbor and Conditional Random Fields. The accuracy of recognition rate is about 82.61 percent.
Robust Speech Recognition Using PCA-Based Noise Classification
This paper proposes a new environmental noise classification using principal component analysis (... more This paper proposes a new environmental noise classification using principal component analysis (PCA) for robust speech recognition. Once the type of noise is identified, speech recognition performance can be enhanced by selecting the identified noise specific acoustic model. The proposed model applies PCA to a set of noise features, and results from PCA are used by a pattern classifier for noise classification. Instead of including both clean and noisy environments in a single classifier, two-step classification is introduced by separating the clean from noisy environments and then identifying the type of noisy environments. The proposed model is evaluated with four types of noise: white, pink, babble, and car from NOISEX-92 and shows a promising result regardless of signal-to-noise ratio (SNR). 1.
Multi Q-Table Q-Learning
2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)
Q-learning is a popular reinforcement learning technique for solving shortest path (STP) problem.... more Q-learning is a popular reinforcement learning technique for solving shortest path (STP) problem. In a maze with multiple sub-tasks such as collecting treasures and avoiding traps, it has been observed that the Q-learning converges to the optimal path. However, the sum of obtained rewards along the path in average is moderate. This paper proposes Multi-Q-Table Q-learning to address a problem of low average sum of rewards. The proposed method constructs a new Q-table whenever a sub-goal is reached. This modification let an agent to learn that the sub-reward is already collect and it can be obtained only once. Our experimental results show that a modified algorithm can achieve an optimal answer to collect all treasures (positive rewards), avoid pit and reach goal with the shortest path. With a small size of maze, the proposed algorithm uses the larger amount of time to achieved optimal solution compared to the conventional Q-learning.
arXiv: Machine Learning, 2018
Counterfactual inference has become a ubiquitous tool in online advertisement, recommendation sys... more Counterfactual inference has become a ubiquitous tool in online advertisement, recommendation systems, medical diagnosis, and finance. Accurate modeling of outcome distributions associated with different interventions---known as counterfactual distributions---is crucial for the success of these applications. In this work, we propose to model counterfactual distributions using a novel Hilbert space representation called counterfactual mean embedding (CME). The CME embeds the associated counterfactual distribution into a reproducing kernel Hilbert space (RKHS) endowed with a positive definite kernel, which allows us to perform causal inference over the entire landscape of the counterfactual distribution. Based on this representation, we propose a distributional treatment effect (DTE) which can quantify the causal effect over entire outcome distributions. Our approach is nonparametric as the CME can be estimated consistently from observational data without requiring any parametric assu...
ArXiv, 2018
This short paper provides further details of the Sloth Search System, which was developed by the ... more This short paper provides further details of the Sloth Search System, which was developed by the NECTEC team for the Video Browser Showdown (VBS) 2018.
A simulated-data adaptation technique for robust speech recognition
Learning with Additional Distributions
This paper studies the problem of learning with distributions. In this work, we do not focus on t... more This paper studies the problem of learning with distributions. In this work, we do not focus on the distribution that represents each data point. Instead, we consider the distribution that is an additional information around each data point. The proposed method yields a new kernel that is similar to an existing one. The main difference is that our kernel requires an integration in the kernel space. Theoretically, the proposed method yields a better generalization compared to normal SVM.
Neural network interpretation of the Parkinson's disease diagnosis from SPECT imaging
Parkinson's disease (PD) diagnosis mainly relies on the visual and semi-quantitative analysis... more Parkinson's disease (PD) diagnosis mainly relies on the visual and semi-quantitative analysis of medical imaging using single-photon emission computed tomography (SPECT) with 123I-Ioflupane (DaTSCAN). The deep learning approach has benefits over other machine learning methods as the model does not rely on feature extraction. However, the complexity of the deep learning model usually results in the difficulty of model interpretation when used in the clinical settings. The model interpretability depends on the interpretation method to reveal the contribution of each pixel in the input image from an attention map. In this paper, we modify the architecture of six well-known interpretation methods to be applicable for 3-dimensional convolutional neural network (3D-CNN) and propose an evaluation method using the Dice coefficient to measure the interpretation performance. The four deep learning models based on the 3D-CNN with high accuracy were applied with our evaluation method. Guide...
Uploads
Papers by Sanparith Marukatat