Abstract:Scene understanding of remote sensing images is of great significance in various applications. Its fundamental problem is how to construct representative features. Various convolutional neural network architectures have been proposed for automatically learning features from images. However, is the current way of configuring the same architecture to learn all the data while ignoring the differences between images the right one? It seems to be contrary to our intuition: it is clear that some images are easier to recognize, and some are harder to recognize. This problem is the gap between the characteristics of the images and the learning features corresponding to specific network structures. Unfortunately, the literature so far lacks an analysis of the two. In this paper, we explore this problem from three aspects: we first build a visual-based evaluation pipeline of scene complexity to characterize the intrinsic differences between images; then, we analyze the relationship between semantic concepts and feature representations, i.e., the scalability and hierarchy of features which the essential elements in CNNs of different architectures, for remote sensing scenes of different complexity; thirdly, we introduce CAM, a visualization method that explains feature learning within neural networks, to analyze the relationship between scenes with different complexity and semantic feature representations. The experimental results show that a complex scene would need deeper and multi-scale features, whereas a simpler scene would need lower and single-scale features. Besides, the complex scene concept is more dependent on the joint semantic representation of multiple objects. Furthermore, we propose the framework of scene complexity prediction for an image and utilize it to design a depth and scale-adaptive model. It achieves higher performance but with fewer parameters than the original model, demonstrating the potential significance of scene complexity.

Simplicity in Complexity : Explaining Visual Complexity using Deep Segmentation Models

Visual Complexity of Shapes: a Hierarchical Perceptual Learning Model

Measuring and Evaluating the Visual Complexity of Chinese Ink Paintings

Understanding Visual Feature Reliance through the Lens of Complexity

Region Intensity Complexity Active Contours.

Scene Complexity: A New Perspective on Understanding the Scene Semantics of Remote Sensing and Designing Image-Adaptive Convolutional Neural Networks

Research on the image complexity based on neural network

A Novel Explainable Artificial Intelligence Model in Image Classification problem

Color Image Complexity versus Over-Segmentation: A Preliminary Study on the Correlation between Complexity Measures and Number of Segments

Modeling the Perception of Visual Complexity in Texture Images

Predicting Complexity Perception of Real World Images

Multi-scale structural complexity as a quantitative measure of visual complexity

Leveraging Image Complexity in Macro-Level Neural Network Design for Medical Image Segmentation

Fusing Multiple Visual Features for Image Complexity Evaluation

A Novel Biologically Inspired Visual Cognition Model: Automatic Extraction of Semantics, Formation of Integrated Concepts, and Reselection Features for Ambiguity

Visual complexity: a review

Principles, applications, and advancements of the Segment Anything Model

Image complexity based fMRI-BOLD visual network categorization across visual datasets using topological descriptors and deep-hybrid learning

Visual Saliency Detection Via Image Complexity Feature

Study on image complexity based on cognitive layering method