Morphological Operation Hsv Color Space Tree Detetction

An automatic tree detection and analyzing method from aerial imagery can aid us in many ways such as keeping track of the number of trees which could be beneficial for forest resource management and others. Given a record of the number of trees in a region can stop deforestation, which is the most arguable issue for every country around the world, therefore, a detailed study of tree count and detection is most required for effective management and quantitative analysis of forest. In this study, we proposed an approach that can automatically segment regions with trees and estimate a tree count on the input image. However, detecting individual tree and counting can be a tough task, and even be inaccurate sometimes. All this depend on the conditions and quality of the image taken. In this study, we propose and compare different approaches for tree detection and counting in a given satellite image.

Our first approach is to apply morphological operations on the image to obtain a clean refined image. Marking local regional minima and maxima on a filtered image can assist in locating crown centroids and marking boundaries. At last a marker-controlled watershed segmentation is applied on the image to separate two touching tree crowns.

Many regions contain spacing between trees including small plants and shrubs which contribute to tree count thereby giving a false count on the number of trees in that region. To remove this ambiguity between small plants and trees a color based segmenting approach was made to discriminate between plants and trees. HSV color space based method is well suited for this purpose as HSV color remove any illumination in an image. After conversion and enhancing the colors we can filter out the small plants and shrubs with their respective hue values compared with those of trees. Therefore, now segmenting and applying watershed transformation will give more accurate tree count in these regions.

Nowadays where Deep Learning has gained a massive popularity over time because of its ability to learn and analyze data at a much faster and accurate way that it is sometimes better than any human being. Research has been conducted on many different general aerial images to automatically label an aerial imagery with specific categories, in recent years researches and numerous algorithms have been developed and implemented for this sole purpose. Many of which include machine learning and deep learning approach. The result from all these shows that deep learning is the best method over satellite imagery dataset.

Aerial imagery of the tree includes only the crown part of the tree which have many irregularities unlike the manmade structure such as buildings, roads that have definite geometry and are easy to identify and classify.

In order to classify individual tree in deep learning approach, we implement a Convolution Neural network (CNN) for this task. The CNN model is trained with two different dataset having different classes of the tree and non-tree images. So that the model can predict the correct result on different tree crowns. The deep leaning image classification model is trained with the help of Matlab with parallel computing toolbox for faster processing and acceleration.