Neural network parameters affecting image classification

Defence Science Journal

K.C. Tiwari¹ 

[1]Army Headquarters, Kashmir House, New Delhi

关键词: Remote sensing;    Digital image classification;    Artificial neural network technique;    Knowledge based classification techniques;    Fuzzy techniques;    Multi band remote sensing data;   

DOI  :  

学科分类：社会科学、人文和艺术（综合）

来源: Defence Scientific Information & Documentation Centre

【 摘 要 】

The study is to assess the behaviour andimpact of various neural network parameters and their effects on theclassification accuracy of remotely sensed images which resulted insuccessful classification of an IRS-1B LISS II image of Roorkee and itssurrounding areas using neural network classification techniques. Themethod can be applied for various defence applications, such as for theidentification of enemy troop concentrations and in logistical planningin deserts by identification of suitable areas for vehicular movement.Five parameters, namely training sample size, number of hidden layers,number of hidden nodes, learning rate and momentum factor were selected.In each case, sets of values were decided based on earlier worksreported. Neural network-based classifications were carried out for asmany as 450 combinations of these parameters. Finally, a graphicalanalysis of the results obtained was carried out to understand therelationship among these parameters. A table of recommended valuesforthese parameters for achieving 90 per cent and higher classificationaccuracy was generated and used in classification of an IRS-1B LISS IIimage. The analysis suggests the existence of an intricate relationshipamong these parameters and calls for a wider series of classificationexperiments as also a more intricate analysis of the relationships.

【 授权许可】

Unknown