KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 7 of 7

Some of the metrics are blocked by your
consent settings
Instance Segmentation of Multiple Myeloma Cells Using Deep-Wise Data Augmentation and Mask R-CNN
(2022-01-01)
Paing, May Phu
;
Sento, Adna
;
Bui, Toan Huy
;
Pintavirooj, Chuchart
Multiple myeloma is a condition of cancer in the bone marrow that can lead to dysfunction of the body and fatal expression in the patient. Manual microscopic analysis of abnormal plasma cells, also known as multiple myeloma cells, is one of the most commonly used diagnostic methods for multiple myeloma. However, as it is a manual process, it consumes too much effort and time. Besides, it has a higher chance of human errors. This paper presents a computer-aided detection and segmentation of myeloma cells from microscopic images of the bone marrow aspiration. Two major contributions are presented in this paper. First, different Mask R-CNN models using different images, including original microscopic images, contrast-enhanced images and stained cell images, are developed to perform instance segmentation of multiple myeloma cells. As a second contribution, a deep-wise augmentation, a deep learning-based data augmentation method, is applied to increase the performance of Mask R-CNN models. Based on the experimental findings, the Mask R-CNN model using contrast-enhanced images combined with the proposed deep-wise data augmentation provides a superior performance compared to other models. It achieves a mean precision of 0.9973, mean recall of 0.8631, and mean intersection over union (IOU) of 0.9062.
Some of the metrics are blocked by your
consent settings
A neural network PID-like controller using a hybrid of online Actor-Critic reinforcement algorithm with the square root cubature Kalman filter
(2018-01-01)
Sento, Adna
;
Kitjaidure, Yuttana
This paper presents a new model of the Neural Network PID-Like controller using an Actor-Critic reinforcement algorithm, called the Neural Network PID-Like controller using an Actor-Critic reinforcement algorithm (NNPID-AC). The proposed NNPID-AC controller is designed to develop the performances and the speed of calculation under the iterative learning algorithm. In the learning algorithm, the critic algorithm receives the reward value and control input to criticize the current state using the action-state value function approximation. Furthermore, instead of applying every available action to predict the local successor state, the algorithm only uses one-step estimation using the fifth degree spherical-radial cubature rule algorithm. To evaluate the proposed NNPID-AC controller, the robot arm MATLAB simulations have been implemented and provide the control system with the load and noise to prove the robustness and fault tolerance, respectively. From the results, the robot arm control system simulation under the control of the proposed NNPID-AC controller can potentially track the error and gives the best responses compared with the other conventional controller either with or without the load and the noise disturbance.
Some of the metrics are blocked by your
consent settings
Forward kinematic-like neural network for solving the 3D reaching inverse kinematics problems
(2017-11-03)
Srisuk, Pannawit
;
Sento, Adna
;
Kitjaidure, Yuttana
This paper presents the inverse kinematic solutions based on neural networks. General neural network approaches use data of the end-effector positions as an input and angle joints as an output to train the neural network for mapping the input to the output. However, the proposed method creates the custom networks from forward kinematic equations. This special structure makes the network like a position finder with ability to automatically adjust angle joints until the end-effector reaches the desired position by backpropagation with variable learning rate algorithm. Then, the solutions of angles can be found from the final weights and bias values. Moreover, the proposed network use less number of neurons and amount of the solution space is not depend on the training data. Finally, to evaluate the performance algorithm, the MATLAB Program is used to demonstrate a 4-DOF robotic arm movement in 3-dimensional. As a result, the proposed algorithm can help a robotic arm move to the desired position (3D reaching) quickly and correctly.
Some of the metrics are blocked by your
consent settings
Inverse kinematics solution using neural networks from forward kinematics equations
(2017-03-23)
Srisuk, Pannawit
;
Sento, Adna
;
Kitjaidure, Yuttana
This paper presents the inverse kinematics solution using the neural network for a robotic arm in 3-dimension. This paper creates neural networks to represent x, y and z position of the end-effector in the forward kinematics equations. The structure of the network has 4 layers; input layer, 2 hidden layers, and output layer. The input and output layers are defined as robotic arm angle and position of the end-effector, respectively. Then, the network updates the weights by the backpropagation with variable learning rate algorithm until reaching criteria that the output of the network is equal to the desired positions. Finally, the inverse kinematics solution is defined by the optimal weights of the network. To evaluate the performance algorithm, the MATLAB Program is used to demonstrate the robotic arm movement in 3-dimension. As a result, the proposed algorithm can help the robotic arm move to the desired position quickly and correctly.
Some of the metrics are blocked by your
consent settings
An intelligent system architecture for meal assistant robotic arm
(2017-03-23)
Sento, Adna
;
Srisuk, Pannawit
;
Kitjaidure, Yuttana
A meal assistant robotic arm is necessary for disable people who cannot move by themselves such as paralysis patients, severely handicapped people, etc. Several research articles have been proposed. However, the system designs for the meal assistant robotic arm have been found for a small number of the improvements. Therefore, this paper presents a detailed study to innovate the meal assistant robotic arm. The proposed system design consists of four parts; 1) feature extraction algorithm using the Microsoft Kinect sensor to create the target position in 3-dimensional Cartesian coordinate, 2) inverse kinematic algorithm to convert the Cartesian coordinate into the joint angles, 3) controller algorithm. The proposed controller uses a new weight updating rule model of the neural network using multi-loop calculation based on the fusion of the gradient algorithm with the cubature Kalman filter (CKF) which can optimize the internal predicted state of the updated weights to improve the proposed controller performances, and 4) the 4-joint robotic arm. To evaluate the performances, the Matlab program is used to implement the overall system. The experimental results show that the meal assistant robotic arm system is able to track the human mouth in the 3-dimensional coordinate system.
Some of the metrics are blocked by your
consent settings
A hybrid CKF-NNPID controller for MIMO nonlinear control system
(2016-09-06)
Sento, Adna
;
Kitjaidure, Yuttana
This paper presents a detailed study to demonstrate the online tuning dynamic neural network PID controller to improve a joint angle position output performance of 4-joint robotic arm. The proposed controller uses a new updating weight rule model of the neural network architecture using multi-loop calculation of the fusion of the gradient algorithm with the cubature Kalman filter (CKF) which can optimize the internal predicted state of the updated weights to improve the proposed controller performances, called a Hybrid CKF-NNIPD controller. To evaluate the proposed controller performances, the demonstration by the Matlab simulation program is used to implement the proposed controller that connects to the 4-joint robotic arm system. In the experimental result, it shows that the proposed controller is a superior control method comparing with the other prior controllers even though the system is under the loading criteria, the proposed controller still potentially tracks the error and gives the best performances.
Some of the metrics are blocked by your
consent settings
Neural network controller based on PID using an extended Kalman filter algorithm for multi-variable non-linear control system
(2016-04-07)
Sento, Adna
;
Kitjaidure, Yuttana
The Proportional Integral Derivative (PID) controller is widely used in the industrial control application, which is only suitable for the single input/single output (SISO) with known-parameters of the linear system. However, many researchers have been proposed the neural network controller based on PID (NNPID) to apply for both of the single and multi-variable control system but the NNPID controller that uses the conventional gradient descent-learning algorithm has many disadvantages such as a low speed of the convergent stability, difficult to set initial values, especially, restriction of the degree of system complexity. Therefore, this paper presents an improvement of recurrent neural network controller based on PID, includeing a controller structure improvement and a modified extended Kalman filter (EKF) learning algorithm for weight update rule, called ENNPID controller. We apply the proposed controller to the dynamic system including inverted pendulum, and DC motor system by the MATLAB simulation. From our experimental results, it shows that the performance of the proposed controller is higher than the other PID-like controllers in terms of fast convergence and fault tolerance that are highly required.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results