برآورد رطوبت خاک مناطق گرمسیری با استفاده از تصاویر رنگی و شبکه های عصبی مصنوعی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|52468||2015||7 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : CATENA, Volume 135, December 2015, Pages 100–106
Information on soil moisture is important for various agricultural, environmental, and hydrological applications; thus, moisture must be determined with the greatest possible accuracy. In this study, based on the fact that soil changes colour as the moisture content changes, artificial neural networks (ANNs) were applied to estimate the moisture content of tropical soils from colour photographs taken with a digital camera. Three different soils were used to train and test the network, and data were collected from disturbed samples subjected to different water contents in the laboratory. MLP (multilayer perceptron) ANNs with one hidden neuron layer and three input variables (red, green, and blue) related to colour were used. To train the networks, various tests were performed by varying the number of hidden layer neurons and using input data of the three soils. The best performing ANN had a hidden layer with twelve neurons and used the tan-sigmoid transfer function. It was found that a single network could estimate the moisture content of all the soils studied from the photographs. The best ANN, which was trained with data of the three soils simultaneously, was also tested with individual soil data separately, and better results were obtained (RMSE ranging from 0.0321 to 0.0650 g/g and r2 ranging from 0.6675 to 0.8231). Although the results are satisfactory, the simplicity of the experiment likely restricted a stronger characterisation of the pattern of soil colour variation due to the change in its moisture content, which thus reduced the performance of the method. However, the proposed method represents an advancement in the indirect estimation of soil moisture content because it has the advantages of being practical, rapid, and non-destructive; requiring relatively low cost; and automating the process in the field.