Abstract

This paper presents MoonPath, an innovative framework for lunar surface analysis that combines hyperspectral imaging with computer vision techniques to create comprehensive hazard maps for rover pathfinding. Building upon the mineral mapping methodology developed by Sivakumar and Neelakantan (2015), our approach integrates multiple data sources from the Chandrayaan-2 mission, including high-resolution optical images, terrain mapping camera data, and hyperspectral data. The system employs advanced image processing, machine learning algorithms, and multi-spectral data fusion to identify and classify potential hazards on the lunar surface. Experimental results demonstrate the effectiveness of our approach in detecting craters, analyzing slopes, and generating detailed hazard maps that can be used for autonomous rover navigation.

1. Introduction

Lunar surface analysis and hazard mapping are critical components of successful space missions, particularly for autonomous rover operations. The challenges of navigating the lunar terrain require sophisticated analysis of multiple data types and robust hazard detection systems. This paper introduces MoonPath, a comprehensive solution that addresses these challenges through the integration of hyperspectral imaging and computer vision techniques.

The need for accurate lunar surface analysis arises from several factors. Safe landing site selection is crucial for mission success, while autonomous rover navigation requires detailed understanding of the terrain. Scientific exploration planning and mission risk assessment further emphasize the importance of comprehensive surface analysis.

2. Related Work

Previous approaches to lunar surface analysis have primarily focused on optical image processing, terrain mapping, crater detection, and slope analysis. Recent advances in hyperspectral imaging have enabled mineral composition analysis, surface feature classification, and terrain characterization.

3. Methodology

Note: Images for the output at different stages

3.1 System Architecture

The MoonPath framework consists of three main components: data preprocessing, feature analysis, and hazard mapping. The system processes multiple data sources to create comprehensive hazard maps.

# Data loading and preprocessing
import rasterio
import cv2
import numpy as np
from rasterio.plot import reshape_as_image

# Load TMC data
rasterdat = rasterio.open(img_01_tif, "r+")
img = rasterdat.read()
img = reshape_as_image(img)
img = cv2.cvtColor(img, cv2.COLOR_BGRA2BGR)

# Load relief data
img_relief = cv2.imread(img_01_relief_tif)
img_relief_gray = cv2.cvtColor(img_relief, cv2.COLOR_BGR2GRAY)

3.2 Slope Analysis and Hazard Detection

The system employs a sophisticated approach to slope analysis and hazard detection using relief data:

# Slope analysis and masking
blur = cv2.GaussianBlur(img_roi, (5,5), 0)
blur_hsv = cv2.cvtColor(blur, cv2.COLOR_BGR2HSV)

# Create mask for high slope areas
lower = np.array([0,0,0], dtype = "uint8")
upper = np.array([105,160,15], dtype = "uint8")
mask = cv2.inRange(blur_hsv, lower, upper)
mask = 255 - mask

# Apply mask and highlight high slope areas
img_roi = cv2.bitwise_and(img_roi, img_roi, mask = mask)
img_roi[mask == 0] = (0, 0, 200)

# Process relief data for slope analysis
mask_relief = cv2.inRange(img_relief_gray, 40, 255)  # slope z factor = 20.000000
blur_relief = cv2.GaussianBlur(mask_relief, (21, 21), 0)
canny_edges_relief = cv2.Canny(blur_relief, 1, 6)
canny_edges_3channel_relief = cv2.cvtColor(canny_edges_relief, cv2.COLOR_GRAY2BGR)
canny_edges_3channel_relief *= np.array((1,1,1),np.uint8)

# Combine slope analysis results
img_roi = cv2.addWeighted(img_roi, 0.9, canny_edges_3channel_relief, 1, 0)

3.3 Ridge Detection

Ridge detection is implemented using the Canny edge algorithm with specific parameters optimized for lunar surface features:

# Ridge detection using Canny edge algorithm
gray1 = cv2.cvtColor(img_roi_copy1, cv2.COLOR_BGR2GRAY)
blur1 = cv2.GaussianBlur(gray1, (45, 45), 0)
canny_edges = cv2.Canny(blur1, 4, 10)
canny_edges_3channel = cv2.cvtColor(canny_edges, cv2.COLOR_GRAY2BGR)
canny_edges_3channel *= np.array((1,1,0),np.uint8)  # Yellow edges for visibility

# Combine with original image
img_roi = cv2.addWeighted(img_roi, 0.7, canny_edges_3channel, 0.5, 0.0, 0.0)

3.4 Crater Detection

The system employs a multi-level crater detection approach using the Hough Circle Transform:

# Multi-level crater filtering
gray = cv2.cvtColor(img_roi, cv2.COLOR_BGR2GRAY)
gray_blurred = cv2.blur(gray, (3, 3))

# Primary crater filter for larger craters
detected_circles = cv2.HoughCircles(canny_edges,
                   cv2.HOUGH_GRADIENT, 1, 20, param1 = 35, 
               param2 = 25.5, minRadius = 0, maxRadius = 45)

# Secondary (finer) crater filter for smaller craters
detected_circles1 = cv2.HoughCircles(gray_blurred,
                   cv2.HOUGH_GRADIENT, 1, 20, param1 = 35.7, 
               param2 = 25.6, minRadius = 0, maxRadius = 27)

# Draw detected craters
if detected_circles is not None:
    detected_circles = np.uint16(np.around(detected_circles))
    for pt in detected_circles[0, :]:
        a, b, r = pt[0], pt[1], pt[2]
        cv2.circle(img_roi, (a, b), int(r/1.5), (0, 255, 0), 2)

OVERLAY OF EDGE, RIDGE AND CRATER MAPPING ON ORIGINAL MAP

3.5 Hyperspectral Analysis

The hyperspectral analysis pipeline processes IIRS data to identify mineral compositions and surface features:

# Hyperspectral data processing
import numpy as np
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans

# Load and preprocess hyperspectral data
def process_hyperspectral_data(iirs_data):
    # Reshape data for PCA
    n_samples, n_bands = iirs_data.shape
    data_2d = iirs_data.reshape(-1, n_bands)
    
    # Apply PCA for dimensionality reduction
    pca = PCA(n_components=0.95)  # Preserve 95% variance
    pca_result = pca.fit_transform(data_2d)
    
    # Extract endmembers using K-Means
    n_endmembers = 7
    kmeans = KMeans(n_clusters=n_endmembers)
    kmeans.fit(pca_result)
    
    # Get endmember spectra
    endmember_indices = np.argmin(kmeans.transform(pca_result), axis=0)
    endmember_spectra = data_2d[endmember_indices]
    
    return endmember_spectra, kmeans.labels_

# Analyze spectral signatures
def analyze_spectral_signatures(endmember_spectra):
    # Calculate spectral indices
    absorption_features = []
    for spectrum in endmember_spectra:
        # Identify absorption bands
        absorption_bands = find_absorption_bands(spectrum)
        absorption_features.append(absorption_bands)
    
    return absorption_features

# Map mineral compositions
def map_mineral_compositions(labels, absorption_features):
    mineral_map = np.zeros_like(labels)
    for i, features in enumerate(absorption_features):
        # Classify based on spectral features
        mineral_type = classify_mineral(features)
        mineral_map[labels == i] = mineral_type
    
    return mineral_map

This hyperspectral analysis pipeline enables the identification of distinct mineral compositions and their spatial distribution across the lunar surface, which is crucial for hazard detection and terrain characterization.

4. Implementation

4.1 Technical Details

The implementation is available as an open-source project at:
GitHub Repository

The system incorporates hyperspectral analysis, hazard mapping, image processing, and visualization tools to provide a comprehensive solution for lunar surface analysis.

4.2 Data Sources

We utilize three specific data sources from the Chandrayaan-2 mission, selected using the South Pole Stereographic projection view from the ISRO's Chandrayaan-2 Map Browse interface (https://chmapbrowse.issdc.gov.in/MapBrowse/):

The Terrain Mapping Camera (TMC) provides high-resolution stereo imaging for terrain analysis and slope mapping. The Imaging Infrared Spectrometer (IIRS) delivers hyperspectral data for mineralogical analysis, while the Optical High Resolution Camera (OHRC) offers very high-resolution panchromatic images for detailed feature detection.

The data selection process involved using the South Pole Stereographic projection view in the Chandrayaan-2 Map Browse interface to identify regions of interest around the lunar south pole. We downloaded corresponding datasets from all three instruments and ensured their temporal and spatial alignment.

5. Results and Analysis

5.1 Feature Detection Results

The system successfully identifies high slope areas, ridge features, and crater boundaries. The endmember analysis reveals seven distinct spectral signatures, each corresponding to different mineral compositions in the lunar surface. These endmembers are used to identify mineralogical variations, map surface composition, and detect potential hazards based on material properties.

5.2 Hazard Mapping Results

The final hazard map combines slope analysis, ridge detection, and crater identification to provide a comprehensive view of potential hazards on the lunar surface.

5.3 Dataset and Methodology

Our analysis utilizes data from the Chandrayaan-2 mission's three primary instruments:

1. Terrain Mapping Camera (TMC):

   • High-resolution stereo imaging
   • Used for slope analysis and terrain mapping
   • South Pole Stereographic projection

2. Imaging Infrared Spectrometer (IIRS):

   • Hyperspectral data for mineralogical analysis
   • Spectral range: 0.8-5.0 μm
   • Used for mineral composition mapping

3. Optical High Resolution Camera (OHRC):

   • Very high-resolution panchromatic images
   • Used for detailed feature detection
   • Resolution: 0.32m

The data selection process involved using the South Pole Stereographic projection view in the Chandrayaan-2 Map Browse interface to identify and download corresponding datasets from all three instruments.

5.4 Analysis Methods

Our analysis pipeline incorporates:

1. Image Processing:

   • Gaussian blur for noise reduction
   • Canny edge detection for ridge identification
   • Slope analysis using relief data

2. Hyperspectral Analysis:

   • PCA for dimensionality reduction
   • K-Means clustering for endmember extraction
   • Spectral signature analysis

3. Hazard Mapping:

   • Integration of multiple data sources
   • Feature combination and visualization
   • Hazard probability assessment

6. Discussion

6.1 Implementation Details

The system uses several key parameters for processing. Gaussian blur kernel sizes of (5,5) are used for general processing, while (21,21) is used for relief analysis. The Canny edge detection employs thresholds of 4-10 for ridges, and slope analysis uses a threshold of 40-255.

6.2 Processing Pipeline

The complete processing pipeline includes data loading and preprocessing, slope analysis and masking, ridge detection, and feature combination and visualization. This integrated approach ensures comprehensive hazard detection and mapping.

6.3 Limitations and Challenges

Current limitations of the system include:

1. Data Limitations:

   • Coverage limited to Chandrayaan-2 mission data
   • Dependence on data availability from ISRO
   • Resolution constraints of available instruments

2. Technical Limitations:

   • Processing requirements for large datasets
   • Integration challenges between different data sources
   • Computational complexity of hyperspectral analysis

3. Methodological Limitations:

   • Assumptions in mineral classification

6.4 Future Directions

Future work will focus on:

1. Technical Improvements:

   • Optimization of processing pipeline
   • Enhanced machine learning integration
   • Improved multi-sensor fusion

2. Feature Enhancements:

   • Additional mineral classification methods
   • Dynamic hazard assessment
   • Autonomous path planning capabilities

7. Conclusion

MoonPath provides a comprehensive solution for lunar surface analysis and hazard mapping. The integration of hyperspectral imaging and computer vision techniques offers significant advantages for space mission planning and execution.

References

1. Chandrayaan-2 Mission Data
2. MoonPath GitHub Repository
3. Sivakumar, V., & Neelakantan, R. (2015). Mineral Mapping of Lunar Highland Region using Moon Mineralogy Mapper (M3) Hyperspectral Data. Journal of Geological Society of India, 86(5), 513-518.

A. Implementation Details

Detailed implementation information is available in the project documentation:
Documentation Link