Rongtai Cai

A visual attention model based on hierarchical spiking neural networks

Based on the information processing functionalities of spiking neurons, hierarchical spiking neural networks are proposed to simulate visual attention. Using spiking neural networks inspired by the visual system, an image can be decomposed into multiple visual image components. Based on specific visual image components and image features, a visual attention system is proposed to extract attention areas according to top–down volition-controlled signals. The hierarchical spiking neural networks are constructed with a conductance-based integrate-and-fire neuron model and a set of specific receptive fields in different levels. The simulation algorithm and properties of the networks are detailed in this paper. Simulation results show that the attention system is able to perform visual attention of objects based on specific image components or features, and a demonstration shows how the attention system can detect a house in a visual image. Using the proposed saliency index, attention areas of interest can be extracted from spike rate maps of multiple visual pathways, such as ON/OFF colour pathways. According to this visual attention principle, the visual image processing system can quickly focus on specific areas while ignoring other areas.

Remembering Key Features of Visual Images Based on Spike Timing Dependent Plasticity of Spiking Neurons

The brain has the powerful capability of remembering key features of images. Based on the principle of spike timing dependent plasticity of spiking neurons and the ON/OFF pathways in the visual system, a spiking neural network is proposed to remember key features of visual images. The simulation results show that the network is capable of remembering key features according to a learning rule based on spike timing dependent plasticity. The principle of the network can be used to explain how a spiking neuron-based system can store the key features of visual images. Furthermore, the network can be applied to spiking neuron based artificial intelligent systems to support the processing visual images.

Development of FPGA Toolbox for Implementation of Spiking Neural Networks

Since more and more new findings and principles of intelligence emerge from neuroscience, spiking neural networks become important topics in artificial intelligence domain. However, as high computational complexity of spiking neural networks it is difficult to implement them efficiently using software simulation. In this paper a new hardware implementation method is proposed. In order to implement spiking neural networks more simply, efficiently and rapidly, a toolbox, which is composed of components of spiking neural networks, is developed for neuroscientists, computer scientists and electronic engineers to implement and simulate spiking neural networks in hardware. Using the toolbox a spiking neural network is easy to implement on a FPGA (Field Programmable Gate Arrays) chip, because the toolbox takes advantages of Xilinx System Generator and works in Mat lab Simulink environment. The graphic user interface enables users easy to design and simulate spiking neural networks on FPGAs and speed up run-time. This paper presents the methodology in development of the toolbox and the examples are used to show its promising application.

Simulation of visual attention using hierarchical spiking neural networks

Based on the information processing functionalities of spiking neurons, a hierarchical spiking neural network model is proposed to simulate visual attention. The network is constructed with a conductance-based integrate-and-fire neuron model and a set of specific receptive fields in different levels. The simulation algorithm and properties of the network are detailed in this paper. Simulation results show that the network is able to perform visual attention to extract objects based on specific image features. Using extraction of horizontal and vertical lines, a demonstration shows how the network can detect a house in a visual image. Using this visual attention principle, many other objects can be extracted by analogy.

Memoryless Polynomial RLS Adaptive Filter for Trajectory Target Tracking

In order to find an effective solution to trajectory target tracking, a memoryless polynomial adaptive filter is proposed in this paper. Unlike Volterra adaptive filter, the proposed memoryless polynomial filter is composed of different monomials, which can fit orbit trajectory very well. Besides, the memoryless polynomial filter can be separated into a linearization filter and a transversal filter. Analogous to linear RLS adaptive filter, a RLS adaptive filter is derived from the memoryless polynomial filter, called Memoryless polynomial RLS adaptive filter (MLPRLS adaptive filter). Experiments show that the proposed filters have better performance than that of normal RLS filters in trajectory tracking.

Extraction of breast cancer areas in mammography using a neural network based on multiple features

Brest cancer is the leading cause of death among women. Early detection and treatment are the keys to reduce breast cancer mortality. Mammography is the most effective method for the early detection at present. In this paper, an approach, which combined multiple feature extraction and a neural network model, is proposed to segment the breast cancer X-ray images. Firstly the visual system inspired model is used to extract the feature information of colors, gray scale, entropy, mean and standard deviation in receptive fields of the input neurons in the network. And then the neural network is trained to segment a breast cancer X-ray image into normal area and cancer area. The experiment results show that the approach is able to extract the cancer area in an X-ray image efficiently. This approach can be applied in automatic diagnosis systems of breast cancer.

Notice of Retraction Simplification of Unscented Kalman Filter for Orbit Object Tracking

Aim at the application of orbit object tracking; we adopt certain simplification technology to UKF (Unscented Kalman Filter), which reduce the computational complexity considerably. The state space equation in orbit object tracking is linear; the sigma sampling in unscented transform can be simplified as a composition-add process; the nonlinear transmit of sigma sampling, state vector, measurement vector and their correlation matrix are simplified by an MSUKF (UKF for Mixing system) algorithm. Experiment result shows that, compare with UKF, the proposed algorithm has the same calculation accuracy with considerable lower computational complexity. The amount of computation of proposed algorithm is only 43.33% of that of the UKF.

Geometrid larvae detection using contour feature

Early detection and prediction of pests in field automatically is an important topic in modern agriculture engineering. In this paper, we design a program that can detect geometrid larvae automatically using image processing technology. Firstly, according to color characterization of larvae, we extract brown larvae candidates from image with background of green leaves. We use morphological operation to correct the candidates regions. Then we use elliptic Fourier transform to represent the contours of larval candidates. The representation is a vector. Thereafter we compare all the possible contour vectors of the larvae with the selected larval contour vector. According to the distance between vectors, we determine whether the contour is a true larval contour. The experimental results show that our method can effectively extract geometrid larvae in tea garden images.

License Plate Extraction Using Spiking Neural Networks

In this paper, we present an algorithm for license plate detection and extraction using spiking neural networks (SNNs). We propose an SNN for the detection of license plate by simulating the color perception principle in human beings’ visual system, where synchronization of spiking trains are employed as a color detection function and used to detect the license plate according to the difference of color in the license plate’s patch and those in the other image patches. By doing so, we can extract those image regions that are likely to be license plates. And then we use another SNN to produce the edge images of these candidates by simulating the receptive field of orientation in human beings’ visual cortex. Finally, we extract the license plate from these candidates according to the texture difference between a real license plate image and the distracters, where the numbers of strokes in image rows are served as cues for the texture difference. The experimental results show that the proposed biological inspired SNNs are valid in the detection and extraction of license plate.

CCD performance model and noise control

In this paper, we describe the performance models of a Charge Coupling Device (CCD), and discuss the issues in the CCD image sensor that deteriorates the CCD imaging performance. We analyze the cause of noise in the CCD imaging sensor, and provide methods to control the CCD noise to improve the CCDs imaging quality. Some simulation results are presented in the paper to show the efficiency of the noise control methods.