Satellite Image Classification Using a Hybrid Manta Ray Foraging Optimization Neural Network

Satellite Image Classification Using a Hybrid Manta Ray Foraging Optimization Neural Network

Accurate satellite image classification powers applications from urban planning and agriculture monitoring to disaster management and climate research. This project introduces a novel metaheuristic-optimized deep learning system for land-cover classification. The Manta Ray Foraging Optimization (MRFO) algorithm — a bio-inspired optimizer modeled on the foraging behavior of manta rays — is applied to automatically tune neural network hyperparameters, including layer configuration, learning rate, and activation settings, overcoming the trial-and-error limitations of manual tuning. The optimized network then classifies multispectral satellite imagery into land-cover classes such as water, forest, urban, agricultural land, and barren terrain. Benchmark comparisons against standard CNN and other nature-inspired optimization methods (PSO, GWO) demonstrate improved classification accuracy and faster convergence. The system includes preprocessing with band normalization, data augmentation, and an interactive GIS-style visualization dashboard for classified map outputs. Built with Python, TensorFlow, and geospatial libraries (GDAL, Rasterio), this project is an outstanding remote sensing machine learning project for geoinformatics students, GIS professionals, and researchers in AI-based satellite image analysis and deep learning earth observation applications.

Components

Python 3.8+
TensorFlow
MRFO optimizer module
GDAL & Rasterio (geospatial)
NumPy
GIS-style visualization dashboard

Key Features

• Manta Ray Foraging Optimization (MRFO) automatically tunes layer configuration, learning rate and activations
• Classification of multispectral imagery into water, forest, urban, agricultural and barren classes
• Band normalization and data augmentation preprocessing
• Outperforms standard CNN defaults and PSO/GWO-tuned networks in accuracy and convergence
• Interactive GIS-style visualization of classified land-cover maps

Applications

Geoinformatics students, GIS professionals, agriculture monitoring, urban planning and disaster management agencies.