Trainable Finite Element Neural Network and Intelligent Imaging Device for Pattern Recognition of Biological Object

将有限元神经网络(Fenn)应用于温室菊花植株水分状况的非侵入性监测。神经信息处理采用控制微分方程(泊松方程)。利用有限元技术得到了泊松方程的解。卡尔曼滤波被用作FENN的学习算法。作为FENN应用的一个实际例子，FENN提供了一种从温室菊花叶片的数字图像数据(FENN的相关输入)估计其叶水势(FENN的相关输出)的方法。人工智能涉及开发软件系统，这些软件系统能够执行人们认为是智能的工作，如果人类这样做的话。目前，人工智能研究中最热门的领域之一就是神经网络研究。神经网络在控制工程…中的应用已十分普遍更重要的是，植物工厂的先进控制系统应该包括反馈和/或前馈回路，这些反馈和/或前馈回路使用一组不同的传感器从植物生长中获得信息。要实现这样的控制系统，需要开发一种传感系统，包括一种用于植物生长的特殊的感官信息处理系统(Hashimoto and Noami，1992)。从植物工厂中生长的植物中获取多个交互信息给传感系统的设计带来了巨大的挑战。Murase等人(1993)简要报道了将有限元神经网络应用于工厂控制的可能性。在这项研究中，受有限元空间表示的启发，提出了一种递归型神经网络(Williams and Zipser，1989)，它执行一组相关输入变量和相关输出变量之间的非线性映射，从而对有限元神经网络进行了清晰的描述。作为Fenn应用的一个实际例子，利用Fenn建立了一种温室菊叶水势的非侵入性测定方法。植物的叶水势水平是反映植物水分状况的有用指标。目前还没有一种非侵入性测量植物叶水势的方法，传统的多层神经网络是由许多简单的相互连接的处理单元组成的，这些处理单元之间的联系只是数学上的。人工神经网络可以在物理空间中构建。为了设计空间神经网络，可以采用有限元方法。人工神经网络中包含的单个单元可以通过有限元在物理上相互连接，作为概念上传递信息或信号的媒介。元素节点与神经元重合，从而可以在欧几里得空间中构建类似大脑的神经结构。就像传统的递归神经网络一样，Fenn的所有神经元通过传递信息或信号的介质相互连接。本研究表明，在欧氏空间中可以使用有限元构造神经网络。实践证明，FENN可以作为一种人工智能技术，进行直接的数据处理。可以作为感觉单位的芬恩输入单元在维度上都是交互的。研究表明，FENN在工程应用中具有很大的潜力。较少

The finite element neural network (FENN) was appied to a non invasive technique to monitor the plant water status of greenhouse-grown chrysanthemums. The governing differential equation (Poisson's equation) was utilized for neural information processing. The solution of the Poisson's equation was obtained using the finite element technique. A Kalman filter was used as a learuing algorithm of the FENN.It was demonstrated as a practical example of the FENN applications that the FENN provided a means of estimating the leaf water potentials (correlated outputs of the FENN) of a greenhouse-grown chrysanthemum from the digital image data of its leaf (correlated inputs of the FENN).Artificial intelligence is concerned with developing software systems that are capable of performing work that one would describe as intelligent if a human did it. One of the "hottest" AI research areas is currently the neural network research. Applications of neural networks have been prevalent in control engineer … More ing.Advanced control systems for plant factories should include feedback and/or feed-forward loops with information obtained from growing plants using a set of various sensors. Realization of such a control system requires the development of a sensing system including a particular sensory information processing system for plant growth (Hashimoto and Nonami, 1992). The acquisition of multiple interactive information from growing plants in the plant factory poses significant challenges for the design of sensing systems. A possibility of applications of the finite element neural network to a plant factory control was reported in brief by Murase et al (1993).In this research the clear description of the finite element neural network that performs a non-linear mapping between a set of correlated input variables and correlated output variables of interest by a recurrent type neural network (Williams and Zipser, 1989) inspired by the finite element spatial representation is presented. As a practical example of the FENN applications, the FENN is utilized in order to establish a non-invasive determination method for the leaf water potentials of a greenhouse-grown chrysanthemum. The leaf water potential level of a plant is a useful index of plant water status. Currently no method is available for the non-invasive measurement of the leaf water potential of a plant.The conventional multilayred neural network is made up of many simple interconnected processing elements of which the connections are only mathematical. Artificial neural networks can be structured in a physical space. The finite element method can be employed in order to devise a spatial neural network. The individual units contained in artificial neural network can be interconnected physically by finite elements, serving as media that can conduct information or signals conceptually. The element nodes coincide with neurons so that a brain-like neural structure can be constructed in the Euclidian space. The all neurons of the FENN are connected to each other by media that can transmit information or signals just like a conventional recurrent neural network.This research showed that a neural network can be constructed in Euclidian space using finite elements. It was demonstrated that the FENN can serve as an artificial intelligence technique that can perform directing data processing. The FENN input cells that can serve as sensory units are all interactive dimensionally. From this study it can be concluded that the FENN has a great potential in engineering applications. Less