Improving Automatic Image Annotation System to handle Spatial Analysis of Aerial Imagery

Working with aerial photographs and their spatial analysis by geology experts, as well as the development of data science and image analysis algorithms, is not simple. The topic of AIA and how to improve label systems and the convolutional neural network (CNN) Visual Geometry Group (VGG-19 algorithm) is the subject of numerous research articles. This paper will discuss the use of the AIA system and a database of aerial images. In our method, image features are extracted using an enhanced VGG-19, and closest features are used to automatically predict image tags. The suggested method takes into account correlations between labels, between labels and images, and between individual images. A label quantity prediction (LQP) model is also suggested to forecast the amount of tags, which increases the accuracy of tag prediction. In the proposed model, supervision information is introduced by interactive convex hulls and deals with automatic annotation methods for pixel-level aerial photographs of petrol zones [1]. We suggest a convex hull approach for visually picking polygonal (irregular) petrol and no petrol zones in place of the typical rectangle-/regular-shaped regions. Followed by low availability rate and subsequently petrol detection rate (true-positive) as the objectives of petrol zones monitoring systems (false-positive). In order to expedite annotation, and based on k-nearest neighbors, the authors constructed a KD-tree (kNN). The suggested solution debunks the independent and identical distribution (i.i.d) hypothesis on machine learning techniques while also transforming the classic two-class fire detection problem into a multi-class classification problem to accommodate the complex aerial image background. It is also simple to use because it only requires pixel information and ignores any additional supporting components from various color systems. The comparison demonstrates the capability of the proposed pixel-level annotation method to concurrently improve petrol detection rates and lower availability rates.