[Paper Review] Satellite image classification methods and Landsat 5TM bands

Jamshid Tamouk, Nasser Lotfi+1 more|arXiv (Cornell University)|Aug 8, 2013

Remote-Sensing Image Classification5 references17 citations

TL;DR

This study evaluates three satellite image classification methods—parallelepiped, minimum distance, and chain—using Landsat 5TM data, finding the chain method achieves the highest overall accuracy (79%). It further identifies Band 4 as most effective for improving classification accuracy when combined with other bands, offering optimized band combinations for detecting land cover objects.

ABSTRACT

This paper attempts to find the most accurate classification method among parallelepiped, minimum distance and chain methods. Moreover, this study also challenges to find the suitable combination of bands, which can lead to better results in case combinations of bands occur. After comparing these three methods, the chain method over perform the other methods with 79% overall accuracy. Hence, it is more accurate than minimum distance with 67% and parallelepiped with 65%. On the other hand, based on bands features, and also by combining several researchers' findings, a table was created which includes the main objects on the land and the suitable combination of the bands for accurately detecting of landcover objects. During this process, it was observed that band 4 (out of 7 bands of Landsat 5TM) is the band, which can be used for increasing the accuracy of the combined bands in detecting objects on the land.

Motivation & Objective

• To identify the most accurate classification method among parallelepiped, minimum distance, and chain methods for Landsat 5TM satellite imagery.
• To determine the optimal combination of Landsat 5TM bands that enhances classification accuracy for land cover objects.
• To provide a reference table linking land cover features to suitable multispectral band combinations for improved detection.
• To assess the individual contribution of each of the seven Landsat 5TM bands to classification performance, particularly focusing on Band 4.

Proposed method

• The study applies three supervised classification techniques—parallelepiped, minimum distance, and chain—on Landsat 5TM multispectral data.
• Each method classifies pixels based on spectral response patterns, with the chain method using iterative decision rules to refine classification boundaries.
• Band combinations are evaluated by testing various multispectral subsets to determine their impact on overall classification accuracy.
• A reference table is constructed by synthesizing findings from prior research and analyzing band spectral characteristics to match land cover types with optimal band combinations.
• Band 4 (near-infrared) is specifically analyzed for its influence on classification accuracy across different combinations.
• Overall accuracy is computed using a confusion matrix comparing classified results to reference data.

Experimental results

Research questions

• RQ1 Which of the three classification methods—parallelepiped, minimum distance, or chain—yields the highest overall accuracy in classifying Landsat 5TM satellite images?
• RQ2 Which combination of Landsat 5TM bands produces the most accurate classification results for land cover detection?
• RQ3 How does Band 4 contribute to improved classification accuracy when used in combination with other bands?
• RQ4 What are the most effective band combinations for detecting specific land cover objects such as vegetation, water, and urban areas?
• RQ5 Can a standardized band combination table be developed to guide accurate land cover classification using Landsat 5TM data?

Key findings

• The chain method achieved the highest overall classification accuracy at 79%, outperforming the minimum distance method (67%) and the parallelepiped method (65%).
• Band 4 (near-infrared) was identified as the most influential band in enhancing classification accuracy when used in combination with other bands.
• The study developed a reference table linking specific land cover features (e.g., vegetation, water bodies, urban areas) to optimal multispectral band combinations for accurate detection.
• Combining multiple bands improved classification accuracy compared to single-band classification, with Band 4 consistently contributing positively to the results.
• The chain method’s iterative decision process provided better boundary delineation and reduced misclassification compared to the other two methods.
• The results demonstrate that band selection and classification algorithm choice are critical factors in achieving high accuracy in Landsat 5TM image classification.