Brain tumors are a primary factor causing cancer-related deaths globally, and their classification remains a significant research challenge due to the variability in tumor intensity, size, and shape, as well as the similar appearances of different tumor types. Accurate differentiation is further complicated by these factors, making diagnosis difficult even with advanced imaging techniques such as magnetic resonance imaging (MRI). Recent techniques in artificial intelligence (AI), in particular deep learning (DL), have improved the speed and accuracy of medical image analysis, but they still face challenges like overfitting and the need for large annotated datasets. This study addresses these challenges by presenting two approaches for brain tumor classification using MRI images. The first approach involves fine-tuning transfer learning cutting-edge models, including SEResNet, ConvNeXtBase, and ResNet101V2, with global average pooling 2D and dropout layers to minimize overfitting and reduce the need for extensive preprocessing. The second approach leverages the Vision Transformer (ViT), optimized with the AdamW optimizer and extensive data augmentation. Experiments on the BT-Large-4C dataset demonstrate that SEResNet achieves the highest accuracy of 97.96%, surpassing ViT’s 95.4%. These results suggest that fine-tuning and transfer learning models are more effective at addressing the challenges of overfitting and dataset limitations, ultimately outperforming the Vision Transformer and existing state-of-the-art techniques in brain tumor classification.

Mapping land use and land cover (LULC) is essential for comprehending changes in the environment and promoting sustainable planning. To achieve accurate and effective LULC mapping, this work investigates the integration of Geographic Information Systems (GIS) with Machine Learning (ML) methodology. Different types of land covers in the Lucknow district were classified using the Random Forest (RF) algorithm and Landsat satellite images. Since the research area consists of a variety of landforms, there are issues with classification accuracy. These challenges are met by combining supplementary data into the GIS framework and adjusting algorithm parameters like selection of cloud free images and homogeneous training samples. The result demonstrates a net increase of 484.59 km² in built-up areas. A net decrement of 75.44 km² was observed in forest areas. A drastic net decrease of 674.52 km² was observed for wetlands. Most of the wastelands have been converted into urban areas and agricultural land based on their suitability with settlements or crops. The classifications achieved an overall accuracy near 90%. This strategy provides a reliable way to track changes in land cover, supporting resource management, urban planning, and environmental preservation. The results highlight how sophisticated computational methods can enhance the accuracy of LULC evaluations.

Creating a crop type map is a dominant yet complicated model to produce. This study aims to determine the best model to identify the wheat crop in the Haridwar district, Uttarakhand, India, by presenting a novel approach using machine learning techniques for time series data derived from the Sentinel-2 satellite spanned from mid-November to April. The proposed methodology combines the Normalized Difference Vegetation Index (NDVI), satellite bands like red, green, blue, and NIR, feature extraction, and classification algorithms to capture crop growth's temporal dynamics effectively. Three models, Random Forest, Convolutional Neural Networks, and Support Vector Machine, were compared to obtain the start of season (SOS). It is validated and evaluated using the performance metrics. Further, Random Forest stood out as the best model statistically and spatially for phenology parameter extraction with the least RMSE value at 19 days. CNN and Random Forest models were used to classify wheat crops by combining SOS, blue, green, red, NIR bands, and NDVI. Random Forest produces a more accurate wheat map with an accuracy of 69% and 0.5 MeanIoU. It was observed that CNN is not able to distinguish between wheat and other crops. The result revealed that incorporating the Sentinel-2 satellite data bearing a high spatial and temporal resolution with supervised machine-learning models and crop phenology metrics can empower the crop type classification process.

To study the environment of the Kipushi mining locality (LMK), the evolution of its landscape was observed using Landsat images from 2000 to 2020. The evolution of the landscape was generally modified by the unplanned expansion of human settlements, agricultural areas, associated with the increase in firewood collection, carbonization, and exploitation of quarry materials. The problem is that this area has never benefited from change detection studies and the LMK area is very heterogeneous. The objective of the study is to evaluate the performance of classification algorithms and apply change detection to highlight the degradation of the LMK. The first approach concerned the classifications based on the stacking of the analyzed Landsat image bands of 2000 and 2020. And the second method performed the classifications on neo-images derived from concatenations of the spectral indices: Normalized Difference Vegetation Index (NDVI), Normalized Difference Building Index (NDBI) and Normalized Difference Water Index (NDWI). In both cases, the study comparatively examined the performance of five variants of classification algorithms, namely, Maximum Likelihood (ML), Minimum Distance (MD), Neural Network (NN), Parallelepiped (Para) and Spectral Angle Mapper (SAM). The results of the controlled classifications on the stacking of Landsat image bands from 2000 and 2020 were less consistent than those obtained with the index concatenation approach. The Para and DM classification algorithms were less efficient. With their respective Kappa scores ranging from 0.27 (2000 image) to 0.43 (2020 image) for Para and from 0.64 (2000 image) to 0.84 (2020 image) for DM. The results of the SAM classifier were satisfactory for the Kappa score of 0.83 (2000) and 0.88 (2020). The ML and NN were more suitable for the study area. Their respective Kappa scores ranged between 0.91 (image 2000) and 0.99 (image 2020) for the LM algorithm and between 0.95 (image 2000) and 0.96 (image 2020) for the NN algorithm.