Q. 1
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is the primary difference between Two-Stage and One-Stage Object Detectors?
💡
Step-by-Step Explanation & Concept Rationale
Two-stage models prioritize detection accuracy/mAP; one-stage models achieve real-time inference speeds (>60 FPS) suitable for video streams and drone feeds.
Q. 2
Artificial Intelligence
Difficulty: Hard
(1 Mark)
In 'Faster R-CNN', what is the role of the 'Region Proposal Network' (RPN)?
💡
Step-by-Step Explanation & Concept Rationale
RPN replaced slow CPU selective search with fast GPU-based anchor proposals, sharing convolutional features with the downstream detection network.
Q. 3
Artificial Intelligence
Difficulty: Medium
(1 Mark)
In YOLO (You Only Look Once), what is the 'Non-Maximum Suppression' (NMS) post-processing algorithm?
💡
Step-by-Step Explanation & Concept Rationale
NMS cleans up dense multi-box detections, ensuring each unique physical object is localized with exactly one bounding box.
Q. 4
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Intersection over Union' (IoU) / Jaccard Index in object detection and semantic segmentation evaluation?
💡
Step-by-Step Explanation & Concept Rationale
IoU >= 0.5 is the standard threshold for True Positive detection in PASCAL VOC and COCO mAP evaluations.
Q. 5
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Mean Average Precision' (mAP, e.g. mAP@0.5, mAP@0.5:0.95) in object detection?
💡
Step-by-Step Explanation & Concept Rationale
COCO mAP@0.5:0.95 averages AP across 10 IoU thresholds (from 0.50 to 0.95 with step 0.05) to rigorously penalize localization inaccuracies.
Q. 6
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'U-Net' architecture (Ronneberger et al. 2015) and why is it the gold standard for biomedical and satellite image segmentation?
💡
Step-by-Step Explanation & Concept Rationale
U-Net skip connections preserve precise pixel-level spatial boundaries lost during downsampling, essential for flood inundation and tumor boundary segmentation.
Q. 7
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is the primary difference between 'Semantic Segmentation' and 'Instance Segmentation'?
💡
Step-by-Step Explanation & Concept Rationale
Instance segmentation (e.g. Mask R-CNN, SAM) detects individual objects with distinct mask boundaries, whereas semantic segmentation (e.g. DeepLab) groups all instances into one class mask.
Q. 8
Artificial Intelligence
Difficulty: Hard
(1 Mark)
In 'Mask R-CNN', what operation replaced 'RoIPool' to eliminate spatial quantization errors and achieve sub-pixel mask alignment?
💡
Step-by-Step Explanation & Concept Rationale
He et al. showed RoIPool's coordinate rounding introduced substantial spatial misalignments; RoIAlign resolves this with continuous bilinear sampling.
Q. 9
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Vision Transformer' (ViT, Dosovitskiy et al. 2020)?
💡
Step-by-Step Explanation & Concept Rationale
ViT showed that standard Transformer architectures with minimal inductive bias surpass CNNs on large pre-training scales (JFT-300M, ImageNet-21k).
Q. 10
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Swin Transformer' (Hierarchical Vision Transformer using Shifted Windows)?
💡
Step-by-Step Explanation & Concept Rationale
Swin Transformer provides multi-scale hierarchical feature maps and linear computational complexity relative to image size, making it a powerful general backbone for dense prediction.
Q. 11
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Segment Anything Model' (SAM, Kirillov et al. / Meta 2023)?
💡
Step-by-Step Explanation & Concept Rationale
SAM was trained on 1 billion masks (SA-1B), enabling zero-shot generalization to segment any object in natural, medical, or satellite imagery.
Q. 12
Artificial Intelligence
Difficulty: Medium
(1 Mark)
In multispectral satellite remote sensing (e.g. Sentinel-2, Landsat-8), what is the 'Normalized Difference Vegetation Index' (NDVI) formula?
💡
Step-by-Step Explanation & Concept Rationale
Healthy green vegetation absorbs red light (chlorophyll absorption) and strongly reflects Near-Infrared (NIR); NDVI ranges from -1 to +1.
Q. 13
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What index is universally calculated from Near-Infrared (NIR) and Green bands to detect and map standing flood waters and open water bodies from satellite imagery?
💡
Step-by-Step Explanation & Concept Rationale
McFeeters NDWI uses Green and NIR to delineate open water surfaces; water has high reflectance in green and near-total absorption in NIR.
Q. 14
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is the 'Normalized Difference Built-up Index' (NDBI) used for in satellite urban analysis?
💡
Step-by-Step Explanation & Concept Rationale
Built-up concrete and urban materials exhibit higher reflectance in Shortwave Infrared (SWIR) compared to NIR.
Q. 15
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Synthetic Aperture Radar' (SAR) (e.g. Sentinel-1 C-band SAR) and why is it indispensable for flood disaster monitoring in cloudy weather?
💡
Step-by-Step Explanation & Concept Rationale
Optical satellites cannot see the ground through cloud cover during active monsoon storms; SAR microwaves penetrate clouds and detect flood water as dark specular reflectors.
Q. 16
Artificial Intelligence
Difficulty: Hard
(1 Mark)
In SAR flood imagery analysis, why do calm standing flood waters appear as dark (low backscatter) pixels in radar images?
💡
Step-by-Step Explanation & Concept Rationale
Specular reflection directs the radar beam away from the sensor (low dB backscatter), creating sharp contrast against rough, bright land surfaces.
Q. 17
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Change Detection' (e.g. Bi-Temporal Siamese Networks) in satellite disaster damage assessment?
💡
Step-by-Step Explanation & Concept Rationale
Siamese networks extract deep feature representations from registered pre/post image pairs to classify structural collapse and flood inundation.
Q. 18
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Spatial Resolution' in satellite remote sensing (e.g. Sentinel-2 at 10m vs WorldView-3 at 0.3m)?
💡
Step-by-Step Explanation & Concept Rationale
Higher spatial resolution (smaller pixel size in meters) allows detection of smaller physical ground objects (individual tents vs broad regional floods).
Q. 19
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Temporal Resolution' (Revisit Time) in satellite constellations?
💡
Step-by-Step Explanation & Concept Rationale
High temporal resolution (frequent revisits) is critical for tracking dynamic, rapid-onset flash floods and cyclone paths in real time.
Q. 20
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Spectral Resolution' in earth observation sensors?
💡
Step-by-Step Explanation & Concept Rationale
Hyperspectral sensors have hundreds of narrow contiguous bands; multispectral sensors have 10-15 broad bands.
Q. 21
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Radiometric Resolution' in remote sensing sensors (e.g. 8-bit vs 12-bit vs 16-bit)?
💡
Step-by-Step Explanation & Concept Rationale
Higher radiometric resolution allows distinguishing subtle moisture and soil variations in shadow or dark terrain.
Q. 22
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Atmospheric Correction' (e.g. BOA / Level-2A surface reflectance) in optical satellite preprocessing?
💡
Step-by-Step Explanation & Concept Rationale
Atmospheric correction ensures reflectance values are physically comparable across different dates and seasons.
Q. 23
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Cloud Masking' (e.g. Fmask / SCL in Sentinel-2) prior to geospatial AI analysis?
💡
Step-by-Step Explanation & Concept Rationale
Cloud shadows have low reflectance and can be easily confused with floodwater unless masked out.
Q. 24
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Pansharpening' in satellite image processing?
💡
Step-by-Step Explanation & Concept Rationale
Pansharpening blends the high spatial detail of panchromatic sensors with the rich spectral data of multispectral sensors.
Q. 25
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Object-Based Image Analysis' (OBIA) compared to Pixel-Based Classification?
💡
Step-by-Step Explanation & Concept Rationale
OBIA uses algorithms like SLIC or multiresolution segmentation to mimic human visual cognitive interpretation.
Study Stream Progress:
Showing 25 of 82 Questions (30%)
Jump to:
Ready to Test Your Retention & Speed?
Now that you have reviewed the study questions and rationales, test yourself in our interactive 1-by-1 practice engine or take the full official timed mock exam.