In: Wagner W„ Szgkely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
574 
IDENTIFICATION OF BEACH FEATURES/PATTERNS THROUGH ARTIFICIAL 
NEURAL NETWORKS TECHNIQUES USING IKONOS DATA 
A. C. Teodoro 1 * 3, *, H. Gonsalves a , J. Pais-Barbosa 3 , F. Veloso-Gomes b , F. Taveira-Pinto b 
3 CICGE, Faculty of Sciences, University of Porto, Rua do Campo Alegre, 687, Porto, Portugal - (amteodor, 
hemani .goncal ves, j pbarbosa) @ fc. up. pt 
b Faculty of Engineering, University of Porto, Rua Roberto Frias, Porto, Portugal - (vgomes, fpinto)@fe.up.pt 
Commission VI, WG VI/4 
KEY WORDS: Coast, Neural, Recognition, Identification, Geomorphology, IKONOS 
ABSTRACT: 
Evaluation of beach hydromorphological behavior and classification of beach hydroforms and hydromorphologies is a complex 
issue. The main objective of this study is to explore pattern recognition methods to identify coastal features/pattems. One of the best 
known approaches for pattern recognition is artificial neural networks (ANNs). In this study an ANN was applied to an IKONOS 
image in order to classify the beach features/pattems. Based on the knowledge of the coastal features, five classes were defined. The 
most common type of ANN used in remote sensing is the multi-layer perceptron (MLP) which was also chosen for this study. The 
number of nodes in the input layer was determined by the number of input bands - the four IKONOS bands (reflectance values): 
blue, green, red and NIR. The output layer consisted of five binary nodes, one for each class: Sea, Suspended-Sediments, Breaking- 
Zone, Beachface and Beach. The ANNs consisted of one hidden layer, with 10 hidden nodes. The dataset was composed by 13775 
pixels unequally comprising the five previously mentioned classes. The dataset was randomly divided into training (70% of each 
class) and validation subsets (30% of each class). Weights connecting the nodes between each layer are initially randomly assigned 
and adjusted during the learning process in order to minimize the global error. The maximum number of allowed iterations was 300. 
The ANN that had been trained with the training data was applied to the validation data. The ANN presented a very good 
performance, demonstrated by the results of the individual class accuracy and overall accuracy (98.6%). The ANN applied in this 
work have been shown to be useful in the recognition of beach features/patterns. 
1. INTRODUCTION 
Beach morphological classification was mainly established for 
Australian and American microtidal sandy environments. 
Different beach morphologic and classification models were 
presented by several authors (Wright and Short, 1984; 
Sunamura, 1988; Lippmann and Holman, 1990; Short and 
Aagaard, 1993; Masselink and Short, 1993; Masselink and 
Hegge, 1995; Short 1991, 1999 and 2006) based on wave, tidal 
and sediment parameters. 
Parameters related to wave, tidal and sediments diameter are 
usually unavailable or nonexistent for the Portuguese coastal 
zone (Pais-Barbosa, 2007 and Pais-Barbosa et al., 2007). 
Therefore, without these parameters, the morphologic analysis 
of high resolution satellite images seems to be a good approach 
to identify and to classify beach morphologies along the 
Portuguese coast. Based on this concept, two different 
approaches were already established in order to identify, 
measure and classify hydroforms and hydromorphologies. 
The first attempt was developed by Pais-Barbosa et al. (2007 
and 2009). This methodology consisted on the visual analysis of 
vertical aerial photographs datasets in a Geographical 
Information System (GIS) environment. However, there are 
some disadvantages associated to this methodology, such as the 
time consumption, the subjectivity introduced by the operator, 
and the impossibility of evaluating the accuracy of this visual 
analysis. 
Teodoro et al. (2009) presented a new approach where a pixel- 
based classification (supervised or unsupervised) and region- 
based approaches (object-oriented classification) were 
employed. These results were compared with the visual 
identification performed by Pais-Barbosa et al. (2007 and 
2009), showing a good agreement between the visual 
identification and the “automatic” classification (Teodoro et al., 
2009), as illustrated in Figure 1. 
Figure 1. Visual and supervised classifications (parallelepiped 
classifier) overlapping (Teodoro et al., 2009) 
* Corresponding author.