Skip to content
CampusPrep
AD-803 (A) · AI for Remote Sensing/Quick Revision Short Notes

AI for Remote Sensing (AD-803 (A)) - Unit 4 Short Notes

How unit 4 is examined

This unit covers four topics: detecting and tracking objects, detecting change over time, hyperspectral and LiDAR analysis, and multi-modal data fusion. None was asked in the supplied papers, so each is taught with a definition, developed points, formulas and a frame for a 7-mark answer in case it appears this year.

Object detection and tracking in remote sensing imagery

<span style="display:inline-block;padding:.16em .6em;border:1.5px solid currentColor;border-radius:999px;font-size:.68em;font-weight:700;letter-spacing:.06em;text-transform:uppercase;opacity:.75">Not asked since 2022</span>

Definition. <mark>Object detection locates and classifies objects such as ships, aircraft, vehicles and buildings in an image by drawing a bounding box and a class label around each one; tracking links the same object across successive images to follow its motion.</mark>

Diagram.

<figure class="ds-fig" style="margin:1.4rem 0;overflow-x:auto"><svg xmlns="http://www.w3.org/2000/svg" id="dsfig-u4-01" viewBox="0 0 467 252" width="467" height="252" role="img" aria-label="Detection pipeline. Img = satellite image, CNN = feature extractor, Box = bounding-box regression, Cls = class label, Trk = tracker linking detections across frames"><style>#dsfig-u4-01 .e{stroke:#454C5A;stroke-width:1.4;fill:none}#dsfig-u4-01 .e.hi{stroke:#2340B8;stroke-width:2.6}#dsfig-u4-01 .n{fill:#FFFFFF;stroke:#16181D;stroke-width:1.4}#dsfig-u4-01 .n.hi{fill:#E3E9FC;stroke:#2340B8;stroke-width:2.2}#dsfig-u4-01 .n.rb-b{fill:#16181D;stroke:#16181D}#dsfig-u4-01 .n.rb-r{fill:#BD3227;stroke:#BD3227}#dsfig-u4-01 text{font-family:"JetBrains Mono",ui-monospace,Menlo,Consolas,monospace;font-size:13px}#dsfig-u4-01 .t{fill:#16181D;font-weight:500}#dsfig-u4-01 .t.inv{fill:#FFFFFF;font-weight:700}#dsfig-u4-01 .kd{stroke:#16181D;stroke-width:1.2}#dsfig-u4-01 .dot{fill:#16181D}#dsfig-u4-01 .ann{fill:#2340B8;font-size:11px;font-weight:700}#dsfig-u4-01 .lbl{fill:#6F7787;font-family:system-ui,-apple-system,sans-serif;font-size:12px;font-weight:700}#dsfig-u4-01 .ptr{fill:#2340B8;font-size:12px;font-weight:700}#dsfig-u4-01 .ah{fill:#454C5A}#dsfig-u4-01 .ah.hi{fill:#2340B8}#dsfig-u4-01 .wl rect{fill:#FFFFFF;stroke:#DCE0E7}#dsfig-u4-01 .wl .t{font-size:12px;font-weight:700}#dsfig-u4-01 .wl.hi rect{fill:#2340B8;stroke:#2340B8}#dsfig-u4-01 .wl.hi .t{fill:#FFFFFF}html.dark #dsfig-u4-01 .e{stroke:#B1B7C3}html.dark #dsfig-u4-01 .e.hi{stroke:#8FA3FF}html.dark #dsfig-u4-01 .n{fill:#161920;stroke:#E6E8ED}html.dark #dsfig-u4-01 .n.hi{fill:#1E2748;stroke:#8FA3FF}html.dark #dsfig-u4-01 .n.rb-b{fill:#E6E8ED;stroke:#E6E8ED}html.dark #dsfig-u4-01 .n.rb-r{fill:#FF7E71;stroke:#FF7E71}html.dark #dsfig-u4-01 .t{fill:#E6E8ED}html.dark #dsfig-u4-01 .t.inv{fill:#0F1115}html.dark #dsfig-u4-01 .kd{stroke:#E6E8ED}html.dark #dsfig-u4-01 .dot{fill:#E6E8ED}html.dark #dsfig-u4-01 .ann{fill:#8FA3FF}html.dark #dsfig-u4-01 .lbl{fill:#858D9C}html.dark #dsfig-u4-01 .ptr{fill:#8FA3FF}html.dark #dsfig-u4-01 .ah{fill:#B1B7C3}html.dark #dsfig-u4-01 .ah.hi{fill:#8FA3FF}html.dark #dsfig-u4-01 .wl rect{fill:#161920;stroke:#2A2E37}html.dark #dsfig-u4-01 .wl.hi rect{fill:#8FA3FF;stroke:#8FA3FF}html.dark #dsfig-u4-01 .wl.hi .t{fill:#0F1115}</style><defs><marker id="ah10" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="ah" d="M0,1 L9,5 L0,9 z"/></marker><marker id="ahh10" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="ah hi" d="M0,1 L9,5 L0,9 z"/></marker></defs><path class="e" d="M59,126 L148,126" marker-end="url(#ah10)"/><path class="e" d="M184.8,115.5 L280.5,51.6" marker-end="url(#ah10)"/><path class="e" d="M184.8,136.5 L280.5,200.4" marker-end="url(#ah10)"/><path class="e" d="M313.8,50.5 L409.5,114.4" marker-end="url(#ah10)"/><path class="e" d="M313.8,201.5 L409.5,137.6" marker-end="url(#ah10)"/><circle class="n" cx="40" cy="126" r="18"/><text class="t" x="40" y="126" dy=".35em" text-anchor="middle">Img</text><circle class="n" cx="169" cy="126" r="18"/><text class="t" x="169" y="126" dy=".35em" text-anchor="middle">CNN</text><circle class="n" cx="298" cy="40" r="18"/><text class="t" x="298" y="40" dy=".35em" text-anchor="middle">Box</text><circle class="n" cx="298" cy="212" r="18"/><text class="t" x="298" y="212" dy=".35em" text-anchor="middle">Cls</text><circle class="n" cx="427" cy="126" r="18"/><text class="t" x="427" y="126" dy=".35em" text-anchor="middle">Trk</text></svg><figcaption style="font-size:.82em;opacity:.72;margin-top:.45rem">Detection pipeline. Img = satellite image, CNN = feature extractor, Box = bounding-box regression, Cls = class label, Trk = tracker linking detections across frames</figcaption></figure>

Key points.

  1. Object detection answers two questions at once: what the object is (classification) and where it is (localisation by a bounding box).
  2. Modern detectors are deep CNN models, either two-stage (Faster R-CNN proposes regions, then classifies them) or one-stage (YOLO, SSD predict boxes and classes in a single pass, so they are faster).
  3. Remote sensing objects are small, densely packed (parked cars, ships in a port), arbitrarily rotated and seen from above, so oriented bounding boxes and high-resolution inputs are often used.
  4. Training needs many labelled images; benchmark datasets such as DOTA and xView provide labelled aircraft, ships and vehicles, and transfer learning from ImageNet reduces the labelling cost.
  5. Accuracy is judged by intersection over union, $IoU = \dfrac{\text{area of overlap}}{\text{area of union}}$; a detection counts as correct when IoU exceeds a threshold, usually 0.5.
  6. Precision is $TP/(TP+FP)$, recall is $TP/(TP+FN)$, and mean average precision (mAP) averages the area under the precision-recall curve over all classes.
  7. Tracking follows an object over time in satellite video or drone footage by matching detections between frames using a Kalman filter for predicted motion and appearance features for identity.
  8. Applications are ship monitoring, vehicle counting, aircraft detection at airfields, building extraction and wildlife counting.

Example. A predicted box overlaps the true box by 60 units of area, and their union is 100 units, so $IoU = 60/100 = 0.6$, which is above 0.5 and is counted as a true positive.

Answer frame. Open with the definition; draw the pipeline diagram; then develop points 1-2 (task and detector types), 3-4 (why remote sensing is hard), 5-6 (IoU and metrics), 7 (tracking); close with one application line such as ship monitoring.

Pitfall: Do not confuse detection with classification; classification gives one label per image, detection gives a label and a location for every object.

Change detection and time-series analysis

<span style="display:inline-block;padding:.16em .6em;border:1.5px solid currentColor;border-radius:999px;font-size:.68em;font-weight:700;letter-spacing:.06em;text-transform:uppercase;opacity:.75">Not asked since 2022</span>

Definition. <mark>Change detection compares co-registered images of the same area taken at different dates to identify what has changed on the ground; time-series analysis studies a sequence of many dates to find trends, seasonal cycles and abrupt events.</mark>

Diagram.

<figure class="ds-fig" style="margin:1.4rem 0;overflow-x:auto"><svg xmlns="http://www.w3.org/2000/svg" id="dsfig-u4-02" viewBox="0 0 467 252" width="467" height="252" role="img" aria-label="Change detection workflow. T1, T2 = images at two dates, Reg = co-registration and radiometric correction, Cmp = comparison method, Map = change map"><style>#dsfig-u4-02 .e{stroke:#454C5A;stroke-width:1.4;fill:none}#dsfig-u4-02 .e.hi{stroke:#2340B8;stroke-width:2.6}#dsfig-u4-02 .n{fill:#FFFFFF;stroke:#16181D;stroke-width:1.4}#dsfig-u4-02 .n.hi{fill:#E3E9FC;stroke:#2340B8;stroke-width:2.2}#dsfig-u4-02 .n.rb-b{fill:#16181D;stroke:#16181D}#dsfig-u4-02 .n.rb-r{fill:#BD3227;stroke:#BD3227}#dsfig-u4-02 text{font-family:"JetBrains Mono",ui-monospace,Menlo,Consolas,monospace;font-size:13px}#dsfig-u4-02 .t{fill:#16181D;font-weight:500}#dsfig-u4-02 .t.inv{fill:#FFFFFF;font-weight:700}#dsfig-u4-02 .kd{stroke:#16181D;stroke-width:1.2}#dsfig-u4-02 .dot{fill:#16181D}#dsfig-u4-02 .ann{fill:#2340B8;font-size:11px;font-weight:700}#dsfig-u4-02 .lbl{fill:#6F7787;font-family:system-ui,-apple-system,sans-serif;font-size:12px;font-weight:700}#dsfig-u4-02 .ptr{fill:#2340B8;font-size:12px;font-weight:700}#dsfig-u4-02 .ah{fill:#454C5A}#dsfig-u4-02 .ah.hi{fill:#2340B8}#dsfig-u4-02 .wl rect{fill:#FFFFFF;stroke:#DCE0E7}#dsfig-u4-02 .wl .t{font-size:12px;font-weight:700}#dsfig-u4-02 .wl.hi rect{fill:#2340B8;stroke:#2340B8}#dsfig-u4-02 .wl.hi .t{fill:#FFFFFF}html.dark #dsfig-u4-02 .e{stroke:#B1B7C3}html.dark #dsfig-u4-02 .e.hi{stroke:#8FA3FF}html.dark #dsfig-u4-02 .n{fill:#161920;stroke:#E6E8ED}html.dark #dsfig-u4-02 .n.hi{fill:#1E2748;stroke:#8FA3FF}html.dark #dsfig-u4-02 .n.rb-b{fill:#E6E8ED;stroke:#E6E8ED}html.dark #dsfig-u4-02 .n.rb-r{fill:#FF7E71;stroke:#FF7E71}html.dark #dsfig-u4-02 .t{fill:#E6E8ED}html.dark #dsfig-u4-02 .t.inv{fill:#0F1115}html.dark #dsfig-u4-02 .kd{stroke:#E6E8ED}html.dark #dsfig-u4-02 .dot{fill:#E6E8ED}html.dark #dsfig-u4-02 .ann{fill:#8FA3FF}html.dark #dsfig-u4-02 .lbl{fill:#858D9C}html.dark #dsfig-u4-02 .ptr{fill:#8FA3FF}html.dark #dsfig-u4-02 .ah{fill:#B1B7C3}html.dark #dsfig-u4-02 .ah.hi{fill:#8FA3FF}html.dark #dsfig-u4-02 .wl rect{fill:#161920;stroke:#2A2E37}html.dark #dsfig-u4-02 .wl.hi rect{fill:#8FA3FF;stroke:#8FA3FF}html.dark #dsfig-u4-02 .wl.hi .t{fill:#0F1115}</style><defs><marker id="ah11" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="ah" d="M0,1 L9,5 L0,9 z"/></marker><marker id="ahh11" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="ah hi" d="M0,1 L9,5 L0,9 z"/></marker></defs><path class="e" d="M55.8,50.5 L151.5,114.4" marker-end="url(#ah11)"/><path class="e" d="M55.8,201.5 L151.5,137.6" marker-end="url(#ah11)"/><path class="e" d="M188,126 L277,126" marker-end="url(#ah11)"/><path class="e" d="M317,126 L406,126" marker-end="url(#ah11)"/><circle class="n" cx="40" cy="40" r="18"/><text class="t" x="40" y="40" dy=".35em" text-anchor="middle">T1</text><circle class="n" cx="40" cy="212" r="18"/><text class="t" x="40" y="212" dy=".35em" text-anchor="middle">T2</text><circle class="n" cx="169" cy="126" r="18"/><text class="t" x="169" y="126" dy=".35em" text-anchor="middle">Reg</text><circle class="n" cx="298" cy="126" r="18"/><text class="t" x="298" y="126" dy=".35em" text-anchor="middle">Cmp</text><circle class="n" cx="427" cy="126" r="18"/><text class="t" x="427" y="126" dy=".35em" text-anchor="middle">Map</text></svg><figcaption style="font-size:.82em;opacity:.72;margin-top:.45rem">Change detection workflow. T1, T2 = images at two dates, Reg = co-registration and radiometric correction, Cmp = comparison method, Map = change map</figcaption></figure>

Key points.

  1. Change detection needs images of the same place, sensor type and season, because different sun angle or season creates apparent change that is not real.
  2. Images must first be co-registered (aligned to sub-pixel accuracy) and radiometrically corrected; misregistration is the main source of false change.
  3. Image differencing computes $D = X_2 - X_1$ for each pixel and marks pixels whose difference exceeds a threshold as changed.
  4. Image ratioing uses $R = X_2 / X_1$, where a value near 1 means no change and it reduces the effect of illumination differences.
  5. Post-classification comparison classifies each date separately and compares the two maps, which gives a from-to change matrix such as forest to urban.
  6. Deep learning methods use Siamese networks: both dates pass through the same shared-weight CNN and the difference of their features is decoded into a change map.
  7. Time-series analysis uses many dates of an index such as $NDVI = \dfrac{NIR - Red}{NIR + Red}$ to track crop growth, deforestation, urban growth and drought.
  8. Recurrent networks (LSTM) and algorithms such as BFAST and LandTrendr model the series to separate seasonal cycles from real disturbance.

Example. A pixel has NDVI 0.72 in 2020 and 0.15 in 2024; $D = 0.15 - 0.72 = -0.57$, a large fall beyond any sensible threshold, so it is marked as vegetation loss.

Answer frame. Open with the definition; draw the workflow diagram; then develop points 1-2 (preconditions), 3-5 (classical methods), 6 (deep learning), 7-8 (time series); close with an application such as deforestation or urban growth mapping.

Pitfall: Skipping co-registration; small misalignment alone produces false change along every edge.

Hyperspectral and LiDAR data analysis

<span style="display:inline-block;padding:.16em .6em;border:1.5px solid currentColor;border-radius:999px;font-size:.68em;font-weight:700;letter-spacing:.06em;text-transform:uppercase;opacity:.75">Not asked since 2022</span>

Definition. <mark>Hyperspectral imaging records hundreds of narrow, contiguous spectral bands for every pixel, giving a continuous spectrum for material identification; LiDAR (Light Detection and Ranging) is an active laser system that measures distance to give a 3D point cloud of the surface.</mark>

Comparison.

Feature Multispectral Hyperspectral LiDAR
Source Passive sunlight Passive sunlight Active laser pulse
Data 3-10 broad bands 100-250+ narrow bands (about 10 nm) 3D points with x, y, z
Main output Land cover map Material spectrum Elevation, height, structure
Main problem Low spectral detail High dimensionality Irregular, unstructured points

Key points.

  1. Hyperspectral data forms a 3D cube (rows, columns, bands); each pixel holds a full spectrum, so materials such as minerals, crop types and water quality are told apart by their spectral signature.
  2. The large number of correlated bands causes the Hughes effect (curse of dimensionality): with few training samples, accuracy falls as bands increase, so dimensionality is reduced by PCA, MNF or band selection.
  3. Spectral unmixing splits a mixed pixel into pure endmembers and their abundance fractions.
  4. Spectral Angle Mapper matches a pixel spectrum $t$ to a reference $r$ using $\theta = \cos^{-1}\dfrac{t \cdot r}{\lVert t \rVert \lVert r \rVert}$; a small angle means a match, and it is insensitive to brightness.
  5. Deep models used are 1D CNN on spectra, 3D CNN on the cube, and hybrid spectral-spatial networks or transformers.
  6. LiDAR range is $R = \dfrac{c\,t}{2}$, where $t$ is the two-way travel time of the pulse and $c = 3 \times 10^8$ m/s; division by 2 is because the pulse goes to the target and back.
  7. LiDAR gives a DSM (surface including trees and buildings) and a DTM (bare ground), so canopy or building height is $CHM = DSM - DTM$.
  8. Point clouds are classified into ground, vegetation and buildings using deep networks such as PointNet or PointNet++, which work directly on unordered points.

Example. A LiDAR pulse returns after $t = 6.67 \times 10^{-6}$ s, so $R = \dfrac{3 \times 10^8 \times 6.67 \times 10^{-6}}{2} \approx 1000$ m.

Answer frame. Open with both definitions; draw or write the comparison table; then develop hyperspectral points 1-5 (cube, dimensionality, unmixing, SAM, deep models) and LiDAR points 6-8 (range formula, DSM and DTM, point-cloud networks); close with applications such as mineral mapping and forest height.

Pitfall: Forgetting the factor of 2 in the LiDAR range formula.

Fusion of multi-modal remote sensing data

<span style="display:inline-block;padding:.16em .6em;border:1.5px solid currentColor;border-radius:999px;font-size:.68em;font-weight:700;letter-spacing:.06em;text-transform:uppercase;opacity:.75">Not asked since 2022</span>

Definition. <mark>Multi-modal data fusion combines data from different sensors, such as optical, SAR, hyperspectral and LiDAR, so that the result is more accurate and informative than any single source.</mark>

Diagram.

<figure class="ds-fig" style="margin:1.4rem 0;overflow-x:auto"><svg xmlns="http://www.w3.org/2000/svg" id="dsfig-u4-03" viewBox="0 0 442 130" width="442" height="130" role="img" aria-label="tree diagram"><style>#dsfig-u4-03 .e{stroke:#454C5A;stroke-width:1.4;fill:none}#dsfig-u4-03 .e.hi{stroke:#2340B8;stroke-width:2.6}#dsfig-u4-03 .n{fill:#FFFFFF;stroke:#16181D;stroke-width:1.4}#dsfig-u4-03 .n.hi{fill:#E3E9FC;stroke:#2340B8;stroke-width:2.2}#dsfig-u4-03 .n.rb-b{fill:#16181D;stroke:#16181D}#dsfig-u4-03 .n.rb-r{fill:#BD3227;stroke:#BD3227}#dsfig-u4-03 text{font-family:"JetBrains Mono",ui-monospace,Menlo,Consolas,monospace;font-size:13px}#dsfig-u4-03 .t{fill:#16181D;font-weight:500}#dsfig-u4-03 .t.inv{fill:#FFFFFF;font-weight:700}#dsfig-u4-03 .kd{stroke:#16181D;stroke-width:1.2}#dsfig-u4-03 .dot{fill:#16181D}#dsfig-u4-03 .ann{fill:#2340B8;font-size:11px;font-weight:700}#dsfig-u4-03 .lbl{fill:#6F7787;font-family:system-ui,-apple-system,sans-serif;font-size:12px;font-weight:700}#dsfig-u4-03 .ptr{fill:#2340B8;font-size:12px;font-weight:700}#dsfig-u4-03 .ah{fill:#454C5A}#dsfig-u4-03 .ah.hi{fill:#2340B8}#dsfig-u4-03 .wl rect{fill:#FFFFFF;stroke:#DCE0E7}#dsfig-u4-03 .wl .t{font-size:12px;font-weight:700}#dsfig-u4-03 .wl.hi rect{fill:#2340B8;stroke:#2340B8}#dsfig-u4-03 .wl.hi .t{fill:#FFFFFF}html.dark #dsfig-u4-03 .e{stroke:#B1B7C3}html.dark #dsfig-u4-03 .e.hi{stroke:#8FA3FF}html.dark #dsfig-u4-03 .n{fill:#161920;stroke:#E6E8ED}html.dark #dsfig-u4-03 .n.hi{fill:#1E2748;stroke:#8FA3FF}html.dark #dsfig-u4-03 .n.rb-b{fill:#E6E8ED;stroke:#E6E8ED}html.dark #dsfig-u4-03 .n.rb-r{fill:#FF7E71;stroke:#FF7E71}html.dark #dsfig-u4-03 .t{fill:#E6E8ED}html.dark #dsfig-u4-03 .t.inv{fill:#0F1115}html.dark #dsfig-u4-03 .kd{stroke:#E6E8ED}html.dark #dsfig-u4-03 .dot{fill:#E6E8ED}html.dark #dsfig-u4-03 .ann{fill:#8FA3FF}html.dark #dsfig-u4-03 .lbl{fill:#858D9C}html.dark #dsfig-u4-03 .ptr{fill:#8FA3FF}html.dark #dsfig-u4-03 .ah{fill:#B1B7C3}html.dark #dsfig-u4-03 .ah.hi{fill:#8FA3FF}html.dark #dsfig-u4-03 .wl rect{fill:#161920;stroke:#2A2E37}html.dark #dsfig-u4-03 .wl.hi rect{fill:#8FA3FF;stroke:#8FA3FF}html.dark #dsfig-u4-03 .wl.hi .t{fill:#0F1115}</style><defs><marker id="ah12" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="ah" d="M0,1 L9,5 L0,9 z"/></marker><marker id="ahh12" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="ah hi" d="M0,1 L9,5 L0,9 z"/></marker></defs><line class="e" x1="203" y1="37" x2="67" y2="101"/><line class="e" x1="203" y1="37" x2="197" y2="101"/><line class="e" x1="203" y1="37" x2="339" y2="101"/><rect class="n" x="142" y="22" width="122" height="30" rx="8"/><text class="t" x="203" y="37" dy=".35em" text-anchor="middle">Fusion levels</text><rect class="n" x="14" y="86" width="106" height="30" rx="8"/><text class="t" x="67" y="101" dy=".35em" text-anchor="middle">Pixel level</text><rect class="n" x="136" y="86" width="122" height="30" rx="8"/><text class="t" x="197" y="101" dy=".35em" text-anchor="middle">Feature level</text><rect class="n" x="274" y="86" width="130" height="30" rx="8"/><text class="t" x="339" y="101" dy=".35em" text-anchor="middle">Decision level</text></svg></figure>

Key points.

  1. Sensors complement each other: optical imagery gives spectral detail but fails under cloud, SAR sees through cloud and works day and night, hyperspectral gives material identity and LiDAR adds height.
  2. Pixel-level fusion merges raw data, for example pan-sharpening, where a high-resolution panchromatic band sharpens a lower-resolution multispectral image using IHS, Brovey or PCA methods.
  3. Feature-level fusion extracts features (texture, spectral indices, height) from each source and concatenates them into one vector before a single classifier.
  4. Decision-level fusion classifies each source separately and combines the outputs by majority voting or weighted averaging.
  5. Deep learning uses two-branch networks, one encoder per modality, joined by concatenation or attention, and trained end to end.
  6. All data must be co-registered to the same grid and resampled to a common resolution before fusion, or the fused product is misleading.
  7. Typical uses are land cover mapping with optical plus SAR, flood mapping under cloud, and urban or forest mapping with hyperspectral plus LiDAR.
  8. Challenges are different resolutions, different noise (speckle in SAR), missing modalities and the cost of labelled data.

Answer frame. Open with the definition and why one sensor is not enough; draw the three-level diagram; then develop points 2-4 (the levels, with an example each), 5 (deep fusion), 6 (co-registration); close with one application, for example flood mapping using SAR plus optical.

Pitfall: Confusing pixel-level (raw data merged) with decision-level (final labels merged) fusion.

Last-minute revision

  • Object detection = bounding box plus class label; tracking = linking the same object across frames.
  • $IoU$ = overlap area divided by union area; a detection is correct when IoU exceeds 0.5.
  • Detectors: two-stage Faster R-CNN (accurate), one-stage YOLO and SSD (fast).
  • Change detection needs co-registered images of the same area at two or more dates.
  • Differencing $D = X_2 - X_1$ then threshold; ratioing $X_2/X_1$; post-classification gives a from-to matrix.
  • $NDVI = (NIR - Red)/(NIR + Red)$ is the usual time-series index.
  • Hyperspectral = hundreds of narrow contiguous bands, a 3D data cube; the Hughes effect is handled by PCA or MNF.
  • Spectral Angle Mapper: a small angle means a close match.
  • LiDAR range $R = ct/2$; $CHM = DSM - DTM$.
  • Fusion levels: pixel, feature, decision; pan-sharpening is pixel-level.

Memory hooks

  • Detection: "box, label, follow" (localise, classify, track).
  • Change: "two dates, align, subtract, threshold".
  • LiDAR: "light echo, half the trip" (divide by 2).
  • Fusion levels: "Pixel, Feature, Decision" = P-F-D, raw data to final labels.
  • Hughes effect: "too many bands, too few samples".

Coverage checklist

  • Object detection and tracking in remote sensing imagery: definition, pipeline diagram, detectors, IoU, tracking; no past questions.
  • Change detection and time-series analysis: definition, workflow diagram, differencing, post-classification, Siamese networks, NDVI series; no past questions.
  • Hyperspectral and LiDAR data analysis: definition, comparison table, Hughes effect, SAM, range formula, DSM and DTM; no past questions.
  • Fusion of multi-modal remote sensing data: definition, levels diagram, pan-sharpening, deep fusion, co-registration; no past questions.
Go to where you left off?

Quick Add to Notes

Save questions, your own notes and screenshots into notes filed by unit. It takes a free account.

Create free account

Have an account? Log in