Skip to content
View Juanmaherruzo's full-sized avatar
🎯
Focusing
🎯
Focusing
  • Córdoba, España

Block or report Juanmaherruzo

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Juanmaherruzo/README.md

I'm a third-year Forestry Engineering student developing production-grade Python pipelines that bring together UAV photogrammetry, multispectral remote sensing, and machine learning to automate forest inventory. My goal is to close the gap between classical silviculture and modern geospatial data science — bringing operational rigor to a field that still relies heavily on manual sampling, paper-to-spreadsheet workflows, and stratification decisions made from memory and PDFs.

The bottleneck in forestry today isn't data. It's tooling. Every meaningful input — UAV imagery, Sentinel-2 time series, LiDAR, NDVI — is already available, often for free. What's missing are pipelines that someone trained as a forester (not as a software engineer) can actually run, on real datasets, on hardware they own. That's the gap I'm trying to work in.

What I build

Three open-source projects on GitHub anchor my work, all MIT-licensed and written in professional English.

Dehesa-Crown-Segmentation-YOLOv11 is an instance-segmentation system for individual tree crown detection on UAV orthomosaics. It uses YOLOv11 with a producer-consumer inference architecture I designed to eliminate VRAM exhaustion on 4 GB laptop GPUs. The pipeline has been validated on a 14,506-hectare dehesa dataset, detecting 357,185 individual trees with documented precision, recall, and mAP metrics for both box and mask predictions.

sentinel2-spectraldex is a Sentinel-2 Level-2A processing pipeline running entirely on the free Copernicus CDSE API. It handles AOI-driven scene search with cloud-cover filtering, OAuth2 authentication with exponential backoff, selective band download, multi-tile MGRS merging in memory, and computation of six spectral indices (NDVI, SAVI, EVI, NBR, NDRE, NDWI). It includes a validation dashboard and a technical note on why NDRE complements NDVI in closed canopies where NDVI saturates.

annotation-station is a full-stack, privacy-first image annotation platform built with FastAPI, React, and SAM 2.1. It's designed for users who cannot send sensitive imagery to cloud-based labelling services. The system uses a three-level embedding cache to run smoothly on 4 GB VRAM laptop GPUs and exports to YOLO-segmentation, YOLO-detection, and COCO formats. The repository includes measured performance comparisons against Roboflow, CVAT, and Label Studio for the local-first use case.

How I work

Object-oriented Python codebases, no hardcoded paths, automated handling of CRS and resolution mismatches, learning-curve diagnostics for every ML component, and explicit documentation of known limitations. I treat reproducibility, honest reporting of model failure modes, and clean engineering as non-negotiable — even on student projects.

How I'm growing

Alongside my degree, I'm following a structured self-directed roadmap across machine learning, deep learning, computer vision, and geospatial data engineering. The progression covers mathematical foundations, the scikit-learn ecosystem and ensemble methods, PyTorch and CNN architectures, transfer learning, and computer vision applied to remote sensing. The goal is to combine academic forestry training with the technical depth needed to deliver production-grade geospatial tools.

Current focus: operational forest carbon MRV, Sentinel-1 SAR for canopy structure, and self-supervised pretraining for remote sensing.

Where I'm headed

I'm interested in operational remote sensing, forest carbon MRV, adaptive management of Mediterranean woodlands, and the integration of GIS and AI inside forestry consultancies and research groups. I see myself contributing to organisations that are serious about turning satellite and UAV data into decisions on the ground — whether that's a consultancy, a public research institute, or a startup working on natural-capital monitoring.

Background and credentials

Third-year Forestry Engineering student at University of Cordoba (UCO). B2 English (certified by Cambridge), actively working toward C1. Spanish native speaker, based in Córdoba.

I haven't completed my degree yet. I haven't done my master's yet. The titles will come — Master in Forestry Engineering and the Geoforest master's are in my plan, alongside continued autodidact work in AI and ML. In the meantime, what I have is public, verifiable work: three open-source projects with real datasets, real metrics, and real engineering decisions documented for anyone to read.

I believe the asymmetry between where I am formally (third-year student, B2) and what's already in my repositories is the most honest signal I can offer. The repositories speak. The titles will follow.

Open to

  • Internships in geospatial ML, precision forestry, remote sensing, or computer vision (Spain or remote).
  • R&D collaborations with research groups, public institutes, or companies working on operational forest monitoring.
  • Technical conversations with people building at the intersection of forestry, earth observation, and applied AI.
  • Mentorship from senior engineers and researchers — I'm at the stage where good feedback is worth more than gold.

If any of this resonates, the fastest way to evaluate fit is to spend 5 minutes on my pinned GitHub repositories. If you want to talk afterwards, I'd be glad to.

Córdoba, Spain · GitHub: Juanmaherruzo

Pinned Loading

  1. annotation_station annotation_station Public

    Local, privacy-first image annotation platform with SAM 2.1-assisted instance segmentation and object detection. FastAPI + React, runs on laptop GPUs, exports to YOLO and COCO

    Python 3

  2. Dehesa-Crown-Segmentation-YOLOv11 Dehesa-Crown-Segmentation-YOLOv11 Public

    Instance segmentation of individual tree crowns in Mediterranean Dehesa ecosystems using YOLOv11

    Jupyter Notebook 3

  3. sentinel2-spectraldex sentinel2-spectraldex Public

    Sentinel-2 L2A pipeline — AOI-driven scene search, selective band download, and six spectral indices (NDVI, SAVI, EVI, NBR, NDRE, NDWI) via the free Copernicus CDSE API.

    Jupyter Notebook 3

  4. AOVE-price-predictor AOVE-price-predictor Public

    Weekly EVOO origin price predictor — bimodal LSTM + Monte Carlo Dropout · MAE 0.204 €/kg · R² 0.943 · FastAPI · Docker

    Jupyter Notebook