An Modified Genetic Neural Network Learning Algorithm And Application In Precision Agriculture

Neural Networks, Fuzzy Techniques and Genetic Algorithms are theories modeled by biological treatment model to process intelligent information. Among them, the neural network is focused on the brain micro-network structure, through the plenty of complex neuronal connections, through self-organization studying and nonlinear dynamics by the formation of the distribution and parallel to deal with some more difficult linguistically information. Fuzzy technology is about the macro function of the brain what is representative of the linguistry and concept, it methods a form as the top to the end, in accordance with the introduction of the membership function and the parallel string rules, deal with any ambiguous linguistically information. The genetic algorithm is simulating an evolution of biological phenomenon (natural selection, crossover and mutation, etc.), it used to explain a approximated searchable method that expresses the natural evolution mechanisms what was in order to solve some difficult questions quickly and effectively. Fuzzy techniques, neural networks and genetic algorithms are in goals similar but methods differ, so they were interacting and improving by each other very much, also they have been a research focus in recent yearsWith the rapid development of agricultural technology, precise agriculture is an ongoing crop cultivation revolution what is base on the achievements of modern information technology, it is above the orientation and navigation, in charge of the soil characteristics and required of crop growth, also with the each process of crops growth and the delivery of various agricultural materials (such as fertilizers, herbicides, pesticides, seed, etc.), to maximize the soil and crops potential , not only satisfy the needs of crop growth but also reduce agricultural waste materials, thereby reduce the consumption, increase profits and protect the ecological environment quality, so keep the agriculture continuable developing. Precision agriculture technology is an important form of modern agricultural production that on the basis of the modern information technology, biotechnology, engineering technology, such as at the latest achievement all in a series of high-tech developed. it became to be an very important aspect during agricultural information collection, data analysis and the implementation of the decision-making over the agricultural technology. In the a variety of factors of technical implementation, the artificial neural network expert system plays an important role, while many problems have been found in the traditional networks which bring by themselves what cannot be evaded, that is hardly to solve the stability and self-adjustability, moreover, the introduction of fuzzy adaptive resonance networks has been existing some problems about easy-off study redundancy arising in some extent from the network structure. so, according this, the dissertation focused on these issues through the primers such as the parallel genetic algorithm global optimization, thought to carry out a algorithm which trying to improve the neural network, and the results would be applied to precision agriculture soil testing classification.After an investigation in the basic principles, theoretical foundation of genetic, as well as foundation and integration of fuzzy logic and neural network, this passage makes a detail research of a fuzzy neural network learning improvement based on modified adaptive genetic algorithm and network topology optimization method. First of all, according to the self-adaptive genetic algorithm, an effective corresponding improvement in strategy has been found which is about the algorithm searching effect between the crossover probability and mutation probability, both modified,adapt to the automatically change with fitness, make what not equal to zero that the largest individual value in species, then what performant excellently individually modified correspondingly in species, so that they won't be in a status similar to stagnation, thus the algorithm finds out of local optimal solution.This disquisition proposed two stage optimization strategy from the genetic algorithm based on the traditional by fuzzy simplified adaptive resonance theory network. The first phase, fix on a optimal inputting sequence of samples, construct primary group for code string by the sequence of network training model, find out individual evaluation criteria by the network classification accuracy, in order to improve the implementation of network performance that locate the optimal training sequence of samples by the genetic algorithm from network. The second phase, struct individual string by the network weights and class identifier, define the individual evaluation criteria as the network's classification accuracy and number of class identification, construct the fitness function, utilize oneself of the modified slfe-adaptive genetic algorithm, and above the original cross and variant operators, a new genetic operator , cutting operator has been added, criteria of cutting operation as network class identification nodes to classes identifiable nodes accuracy and frequentness of this class identifiable nodes being referred to, redundant node cut. Through the algorithm, the optimum strategy accommodates the problem between network's steadiness and slef-adaptive, compare with the simulate experimentation results, the appliance normative testing data-base, there is effective improvement in class accuracy and structure redundant. At the same time, the learning time of modified neural networks spended more than the traditional SFAM network, ssFAM network, ssGAM network and ssEAM network, analysis shows that much time did spended on the genetic algorithm iterative, this is one of the disadvantages in this learning algorithm.Finally, combined to the subject of collected soil samples data of Jilin province Lishu city, take the advantage of the modified neural networks, classify the sample testing data above-mentioned, the classification has achieved satisfactory results. Through a number of experiments showed that the modified neural networks bring forward a new approach in artificial neural networks applicable in traditional agriculture and it explored an expansive applicable foreground in traditional agriculture.