Knowledge extraction and understanding of images using Fuzzy inference neural network

取得了以下研究成果。(1)基于模糊推理神经网络的图像识别与理解在本研究中，提出了一种新的用于图像识别与理解的神经网络系统。它由学习能力强的神经网络和善于处理规则的模糊系统组成。该系统主要包括图像分割、图像识别和图像判读三个过程。通过计算机实验，对71.9%的像素点进行了正确识别，得到了合理的理解结果。(2)考虑相对位置的景物图像识别在识别图像时，物体的相对位置起着重要的作用。提出了考虑实际情况的识别系统。(3)基于假设和检验的风景图像识别系统能够反映图像中的感性信息，并进行假设和检验，以获得更好的识别效果。(4)基于视觉系统的运动目标识别神经网络受生物视觉系统的启发，提出了一种新的神经网络系统。它可以从运动和形状中识别物体。(5)层次并行神经网络在三维物体识别中的应用本文提出了一种新的神经网络，它具有类Neo-cognitron结构和双向信息处理功能。(6)混沌模拟联想记忆，记忆功能对于更高层次的图像处理至关重要。本研究旨在为下一代先进的识别系统。(7)神经网络联想记忆的概念，我们更高层次的智能处理，如思维和推理是基于语言。对于下一步的图像理解，神经网络的语言处理是不可或缺的。在本研究中，我们提出了这样一种神经网络。

The following research results were obtained.(1) Image recognition and understanding using fuzzy inference neural networkIn this research, a new neural network system for image recognition and understanding is proposed. It consists of neural network which has high learning ability and fuzzy system which is good at treating rules. There are 3 processes in the system: segmentation, image recognition, and image interpretation. According to the computer experiments, 71.9% of pixels are correctly recognized and reasonable understanding results are obtained.(2) Scenery image recognition considering relative positionWhen we recognize images, relative positions of objects play an important role. The recognition system considering the fact is proposed.(3) Scenery image recognition based on hypothesis and testingThe recognition system can reflect Kansei information from the images and perform hypothesis and testing for better performance.(4) Moving object recognition neural network based on visual systemA new neural network system inspired by biological visual system is proposed. It can recognize objects from the movement and the shape.(5) Hierarchical-parallel neural network for 3-D object recognitionA new neural network composed of Neo-cognitron like structure and bi-directional information processing is proposed.(6) Chaos analog associative memory.Memory function is crucial for higher level image processing. This research is aimed for advanced recognition system for the next generation.(7) Neural network associative memory for concepts of sentenceOur higher level intelligent processing such as thinking and inference are based on language. For the next step of image understanding, language processing by neural network is indispensable. In this research, we proposed such a kind of neural network.