Digital Image Processing of Remote-Sensed Data
Digital Image Segmentation
Image segmentation in remote sensing is the process of dividing an image into spatially continuous, meaningful regions (segments or objects) that share similar spectral, spatial, or textural characteristics. This step is especially crucial in high-resolution imagery and object-based image analysis (OBIA), where the goal is to interpret real-world features like fields, buildings, roads, or tree crowns. It involves dividing an image into meaningful regions or segments, with pixels belonging to the same segment sharing similar characteristics (like color, texture, or spatial location).
Image Segmentation Techniques
Pixel-Based Image Segmentation
Pixel-based image segmentation is a technique that classifies each pixel in an image into a specific class or category based on its characteristics. This process assigns a label to every pixel, effectively dividing the image into distinct regions or segments. These segments can represent objects, parts of objects, or other meaningful regions within the image. The process of dividing an image into multiple segments, relies on two fundamental pixel-based properties: discontinuity and similarity. Discontinuity-based methods focus on finding abrupt changes in pixel values (like edges) to delineate region boundaries. Conversely, similarity-based methods group pixels with similar characteristics (intensity, color, texture) into regions.
Object-Based Image Segmentation
Object-based image segmentation (often part of object-based image analysis, or OBIA) is a technique where an image is segmented into meaningful objects or groups of pixels based on their spectral, spatial, and contextual properties—not just individual pixels. This differs from pixel-based methods that classify each pixel independently. These objects are then used as the basis for further analysis, such as object-based image classification. Unlike pixel-based approaches, it recognizes that crops and land parcels are spatially connected features—not just isolated pixels. It is best suited for high-resolution imagery
Machine Learning-Based Image Segmentation
Machine learning methods learn patterns from labeled data using features from either pixels or objects. Machine learning (ML)–based image segmentation involves training algorithms to automatically identify and separate regions (objects, crops, zones, etc.) within an image based on patterns learned from data. Machine learning algorithms can be used to classify pixels or objects into different categories (e.g., land cover types like forests, water bodies, urban areas). Deep learning (DL) techniques, a subfield of ML, have gained prominence for remote sensing image segmentation due to their ability to learn complex patterns from data.
Deep Learning-Based Image Segmentation
Deep learning is the state-of-the-art approach that uses convolutional neural networks (CNNs) to classify every pixel in an image. It's especially powerful for high-resolution, complex, and detailed tasks in remote sensing, agriculture, and autonomous systems. In general, there are two main types of deep learning segmentation for agricultural applications.
Semantic Segmentation. Semantic segmentation is one of the basic tasks in machine vision to achieve pixel-level classification. It is an important component of computer vision-based applications. Unlike making predictions about an image, semantic segmentation generates pixel-level descriptions of objects embedded in their spatial information. With the advancement of semantic segmentation methods, they have been used to address diversity and data-rich agricultural problems, such as crop cover and type analysis, crop species labeling, weed segmentation, predictive agriculture, pest and disease identification, etc.
Instance Segmentation. Instance segmentation is a computer vision task that involves identifying and separating individual objects within an image, including detecting the boundaries of each object and assigning a unique label to each object.
Click on the following topics for more information on digital image processing of remote-sensed data.
