New opportunity

Image Processing Engineer

Recrew · Bangalore, India

About this role

This full-time engineering role in Bangalore focuses on astronomical image processing for ground-based telescopes and satellite navigation systems. The engineer will develop machine learning and deep learning models for star tracking and celestial object detection and classification, supporting Attitude Determination and Control System (ADCS) components.

Responsibilities

Design models that identify stars as spots or blobs and other objects as streaks or lines in astronomical images.
Build detection and classification algorithms that distinguish these features and retrieve coordinates with sub-pixel precision.
Train and test models on synthetic and real astronomical images; evaluate and document performance.
Optimize algorithms for real-sky signal and noise conditions, collaborate with hardware teams on ADCS integration, and deliver technical documentation and briefings.

Required qualifications

A bachelor’s or master’s degree in Electronics, Physics, Data Science, Astronomy, or a related technical field.
1–6 years of experience with image processing algorithms and machine learning; fresh graduates are also welcome.
Strong Python and object-oriented programming skills; academic or project experience with CNN architectures for detection and classification.
Hands-on PyTorch and TensorFlow/Keras experience through coursework or personal projects; familiarity with Astropy, Scikit-learn, and SciPy. Strong analytical, problem-solving, and independent learning skills are required.

Preferred qualifications

Understanding of astronomical imagery and real-sky signal and noise; knowledge of Bayesian methods and probabilistic orientation-estimation models.
OpenCV image preprocessing experience, Git and version-control practices, or aerospace, astronomy, or satellite-imaging projects or internships.

The work supports a venture-backed space-component startup serving low-Earth-orbit satellite manufacturers. The posting highlights collaboration with its engineering team and exposure to its US and India operations.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Image processingComputer visionMachine learningDeep learningObject detectionImage classificationConvolutional neural networksPythonObject-oriented programmingPyTorchTensorFlowKerasAstropyScikit-learnSciPySub-pixel localizationModel evaluationAnalytical problem-solvingBayesian methodsProbabilistic modelingOpenCVImage preprocessingGitTechnical documentation

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Distinguish stars from streaking objects
Posting: Develop ML/DL detection and classification models that identify stars as spots or blobs and other objects as streaks or lines. Prepare to explain a CNN-based approach and how it would distinguish the two.
Retrieve precise coordinates
Posting: Algorithms must report object coordinates with sub-pixel precision, with performance evaluated and documented. Prepare to describe how you would measure localization error separately from detection and classification performance.
Test across synthetic and real imagery
Posting: Train and test on both synthetic and real astronomical images, then optimize for real-sky signal and noise conditions. Prepare to explain a validation plan that checks whether results transfer between the two.
Implement maintainable image-processing models
Posting: The role calls for Python and object-oriented implementation, CNN project experience, hands-on PyTorch and TensorFlow/Keras, and familiarity with Astropy, Scikit-learn, and SciPy. Prepare to discuss concrete implementation and testing choices rather than reciting tools.
Support ADCS integration and communicate results
Posting: Collaborate with hardware teams to integrate image processing with ADCS components, and deliver technical documentation and performance briefings. Prepare to explain what coordinate outputs, evaluation results, and implementation details you would communicate.

Likely challenges to think through

Classification versus localization
Possible challenge, not a confirmed employer problem: A model may label a faint blob or streak correctly yet locate it poorly. Prepare to discuss separate metrics and error analysis for class decisions and sub-pixel coordinates.
Synthetic-to-real transfer
Possible challenge, not a confirmed employer problem: Synthetic images can support training, but their signal and noise may not represent real-sky images. Prepare to discuss testing on both datasets and identifying conditions where performance diverges.
Algorithm outputs versus hardware needs
Possible challenge, not a confirmed employer problem: A strong image-processing result may still require changes before ADCS integration. Prepare to discuss how you would agree on coordinate definitions and output formats with hardware engineers, then document the resulting interface and measured performance.

Practice your answers

Technical practice

How would you distinguish stellar blobs from object streaks in an astronomical image?

What to coverThe role calls for detecting and classifying these two feature types. Suggested preparation: explain useful shape and intensity cues, where preprocessing might help, and how you would handle faint or overlapping features.

How would you approach a CNN model for astronomical object detection and classification?

What to coverThe posting asks for academic or project experience with CNN architectures. Suggested preparation: describe a plausible model design, its inputs and outputs, and the trade-off between detection quality and computational cost.

How would you retrieve object coordinates with sub-pixel precision?

What to coverSub-pixel coordinate retrieval is an explicit responsibility. Suggested preparation: compare an initial pixel-level detection with a refined position estimate, and explain how you would measure localization error.

How would you use synthetic and real images to train and test a model?

What to coverThe role involves both dataset types. Suggested preparation: discuss their different strengths, the risk that synthetic images fail to represent real-sky conditions, and how you would keep test data separate from training data.

What would you report when evaluating a detection and classification algorithm?

What to coverThe posting requires documented performance evaluation and sub-pixel coordinate reporting. Suggested preparation: address missed detections, false detections, classification errors, localization error, and how results vary across image conditions.

How would you troubleshoot performance under changing real-sky signal and noise conditions?

What to coverOptimizing for varied signal and noise environments is a stated duty. Suggested preparation: explain how you would isolate a failure mode, compare preprocessing or model changes, and check that an improvement does not harm other conditions.

How would you structure a Python image-processing pipeline?

What to coverThe posting requires Python and object-oriented programming skills and names Astropy, Scikit-learn, and SciPy as relevant libraries. Suggested preparation: describe clear components for image input, preprocessing, inference, and evaluation, with testable interfaces rather than assuming any mandated library workflow.

How would you make model experiments comparable in PyTorch and TensorFlow/Keras?

What to coverHands-on experience with both frameworks is listed. Suggested preparation: explain how you would hold data splits, preprocessing, metrics, and training settings consistent, while noting that implementation differences can affect results.

What would you clarify before integrating image-processing output with an ADCS component?

What to coverThe role includes collaboration with hardware teams on ADCS integration. Suggested preparation: discuss coordinate conventions, output confidence, timing, and failure handling as questions to resolve with the team, not as stated interface requirements.

How might probabilistic methods help with orientation estimation?

What to coverBayesian methods and probabilistic orientation-estimation models are listed as nice-to-have knowledge. Suggested preparation: explain how uncertainty in image-derived observations could be represented and why a probabilistic estimate may be preferable when detections are ambiguous.

Behavioural practice

Tell me about a time you worked with another discipline to integrate a technical solution.

What to coverThe role involves collaboration with hardware teams. Suggested preparation: use a coursework, project, or work example and cover the situation, what you did to resolve an interface or constraint, and the result.

Describe a time you investigated an unexpected model or image-processing result.

What to coverThe posting emphasizes analytical problem-solving and optimization for varied image conditions. Suggested preparation: give the situation, the tests you chose and actions you took, and what the evidence showed in the end.

Tell me about a time you learned a technical method independently to complete a task.

What to coverIndependent learning is a stated criterion. Suggested preparation: identify the situation and learning goal, the steps you took to apply and verify the method, and the resulting outcome.

Describe a time you communicated technical performance results to others.

What to coverThe role requires documentation and briefings on algorithm performance and implementation. Suggested preparation: explain the situation, how you presented evidence and limitations for your audience, and the result of that communication.

Tell me about a time you chose between improving accuracy and meeting a practical constraint.

What to coverThe posting calls for high-performance models and collaboration on hardware integration. Suggested preparation: use a real project or coursework situation, explain the options and your action, and state the measured result or what remained uncertain.

Sharpen your knowledge

Topics to prepare

Stars, streaks and detection errors
Compare sketches of a compact star blob and an elongated object streak; list cases where noise could make them hard to distinguish. The role explicitly calls for detecting and classifying these features in astronomical images.
CNN detection and classification
Sketch how a CNN could produce object classes and locations, then explain how you would test missed stars and misclassified streaks. CNN project or academic experience is a must-have criterion.
Sub-pixel coordinates
Work through a small pixel-intensity example and explain why the brightest pixel alone may not give a sufficiently precise coordinate. The posting explicitly requires sub-pixel coordinate retrieval and performance reporting.
Real-sky signal and noise
Compare a clean synthetic image with a noisier version and predict which detections may fail. Training on synthetic and real images and optimizing for real-sky conditions are stated responsibilities.
Star tracking and ADCS context
Draw a simple flow from image detection to star coordinates to an ADCS component, and identify what information a hardware collaborator would need. ADCS integration is explicit; reviewing Bayesian orientation estimation is optional because the posting lists it as nice to have.

Skills to revise

Python and object-oriented implementation
Write a small Python class that accepts an image and returns detected objects with classes and coordinates; test its behavior on an empty image. Strong Python and OOP implementation skills are must-have criteria.
PyTorch and TensorFlow/Keras practice
Implement and train a small CNN classification example in each framework, then compare validation results and explain the data flow. Hands-on experience with both is explicitly requested through coursework or personal projects.
Blob and streak localization
On a small test image, identify compact and elongated features and calculate a refined coordinate for one detection; report its error against a known position. Detection, classification and sub-pixel precision are stated tasks.
Scientific Python workflow
Revisit a short example using Astropy to handle astronomical image data, SciPy for numerical analysis and Scikit-learn for evaluation; explain each library's role. Familiarity with all three is listed as a must-have criterion.
Synthetic-versus-real evaluation
Plan a train/test comparison for synthetic and real images, record detection and classification errors, and note how performance changes with signal and noise. The role calls for both dataset types, evaluation and documentation.
Technical handoff
Draft a one-page briefing that states coordinate outputs, observed errors, test conditions and open integration questions. The posting calls for technical briefings and collaboration with hardware teams on ADCS integration; the page does not prescribe a particular handoff format.

Your preparation checklist

Select a relevant project walkthrough
Prepare a short walkthrough of an image-processing or CNN detection/classification project, if you have one. Explain your contribution, Python design, training choices, results and limitations; the posting accepts academic and personal projects.
Sketch a detection and localization approach
As a practice exercise, diagram a pipeline that separates star spots or blobs from object streaks or lines and returns sub-pixel coordinates. Be ready to explain how you would test each stage.
Prepare an evaluation example
The role uses synthetic and real astronomical images and documents performance. Prepare a sample results table covering detection, classification and coordinate error, with a plan for checking performance under different signal and noise conditions.
Map your evidence to the stated tools
List specific coursework or project examples for PyTorch, TensorFlow/Keras, CNNs, Python OOP, Astropy, Scikit-learn and SciPy. Note honestly where your experience is limited; OpenCV, Git and probabilistic orientation methods are listed as preferred.
Rehearse an ADCS integration discussion
Prepare questions about image inputs, coordinate outputs, performance constraints and how the algorithm would be tested with hardware. These are discussion prompts, not specified interface requirements; the posting says the engineer collaborates with hardware teams on ADCS integration.
Write questions for the engineering team
Bring a short list asking how synthetic and real-sky data are used, what sub-pixel performance matters most, and how algorithm results are documented and briefed to the team.

Availability last checked: 4 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…