Connecting Data and Intelligence: The Role of FME in Machine Learning

Connecting Data
and Intelligence:
The Role of FME in Machine Learning
The Peak of Data
and AI 2025

2025
The
Peak
of
Data
and
AI
Thiago Brigagão
Director
Danilo de
Lima
Director

1. Solutial
2. From Paper to Place
3. Object Identification
4. Results
Agenda

2025
The
Peak
of
Data
and
AI
Solutial – Soluções e Análise de Dados
We are located in São José dos Campos - SP, and has over 18 years
of market experience. We are the official representatives of the
FME Platform in Brazil and South America.
Complete range of services, including specialized consulting,
project development, and customized training. We hold all the
necessary certifications, provided by Safe Software, to ensure
excellence in each of these activities.

2025
The
Peak
of
Data
and
AI
Market-leading solutions

2025
The
Peak
of
Data
and
AI
Segments
Agriculture &
Forest
Energy GIS & IT Government
Mining Water Oil & Gas
Telco

From Paper to Place:
Extracting Land Intelligence from Old Records

2025
The
Peak
of
Data
and
AI
Imagine…
You need to extract key insights from an 80-
page document — written 20 years ago and
available only as a scanned image.

2025
The
Peak
of
Data
and
AI
Sounds Challenging?

2025
The
Peak
of
Data
and
AI
For many organizations,
this is part of daily life — they need to
make important decisions using old and
hard-to-find information.

2025
The
Peak
of
Data
and
AI
Introduction
Every day, organizations in different areas
need to read land documents to find
important information like the size of the
property, who owns it now, and other legal
or land details.

2025
The
Peak
of
Data
and
AI
Doing this work normally needs a special
team that only reads and checks the
documents.
These documents are usually very
long.

2025
The
Peak
of
Data
and
AI
This means the process is manual, takes
a lot of time and effort, and costs a lot of
money.

2025
The
Peak
of
Data
and
AI
Technical Challenges
1. IMAGE-BASED PDFS
Most documents are scanned images, not text-based files.
2. COMPLEX LAYOUTS
Text is scattered, mixed with notary stamps and visual elements.
3. INCONSISTENT QUALITY
Document clarity and resolution vary significantly.
4. TEXT EXTRACTION
How to accurately read and extract content from image-based documents?
5. TEXT OPTIMIZATION
How to clean and improve the quality of extracted text, especially from old or noisy documents?
6. INFORMATION STRUCTURING
How to automatically generate summaries and extract key data like property owners and land area?

2025
The
Peak
of
Data
and
AI
1. IMAGE-BASED PDFS
Most documents are scanned images, not text-based files.
2. COMPLEX LAYOUTS
Text is scattered, mixed with notary stamps and visual
elements.

2
Extracting the text
Uses Tesseract OCR to identify text in the
processed image.
4
Sending to ChatGPT to
analyze the text
Text into parts and sent to ChatGPT
for step-by-step analysis, enabling
more accurate extraction
z
Applies statistics to recover the average
of the boxes with text.
Applies dissolve to merge the boxes that
are touching.
Recovers the areas of higher text clustering,
excluding other areas that are not of
interest.
Crops the image from the PDF based on the
identified text areas.
Raster processing, noise removal, and
extraction of letter bounding boxes.
Raster processing and text
area identification
PDF Reading (RASTER)
1
Prepares the questions for assembling the
request to ChatGPT.
Separates the fragmented texts and
prepares them for sending to ChatGPT.
Word count to fragment the text into
pieces.
Preparing the text for the OpenAI API
request - ChatGPT
3
5
Sends the response via email to the
requesting user.
Processing the response and sending it via
e-mail to the requester.
Prepares the response and the HTML
template to send the reply via email.

2025
The
Peak
of
Data
and
AI
Key Point
It’s important to highlight that, although the Transformer
OpenAIChatGPTConnector exists, we choose to work directly
with Python code to fully customize the solution according to
the specific needs of the project.
Another key point is the use of the Tiktoken library, which
allows us to track the number of tokens used during both the
input and output phases of the interaction

2025
The
Peak
of
Data
and
AI
Turning lost information into
intelligent decisions is the
true power of automation
combined with artificial
intelligence
— ChatGPT, your co-pilot in data innovation

Object Identification:
Extraction of objects and Data Quality
In development

2025
The
Peak
of
Data
and
AI
One of the largest Energy Distributors in South America.
Total number of customers: around 8.8 million.
Population served: approximately 18 million people.
Coverage area: 785 municipalities in Brazil.
Transmission: Operates more than 10,000 km of transmission lines.

2025
The
Peak
of
Data
and
AI
Our eletrical client has a manual process for
identifying objects, wich is very expensive.

2025
The
Peak
of
Data
and
AI
Drone images over the transmission line are generated, and it is necessary to manually inspect these
images.

2025
The
Peak
of
Data
and
AI
1. RECEIVING THESE DAILY IMAGES
2. OBJECT IDENTIFICATIONS (VISUALLY)
Transformers — to step down or step up electrical voltage.
Fuse cutouts — for circuit protection and disconnection.
Reclosers (automatic circuit reclosers) — to automatically restore power after minor faults.
Insulator bolts / pin or suspension insulators — to hold the wires without energy loss.
Conductor cables — for medium or low voltage.
Crossarms and hardware — for mechanical support.
Smart meters — in advanced urban areas.
Telecommunication boxes — for internet, telephone, and fiber optic cables.
Public lighting (light arms and luminaires) — installed on the same poles.
Grounding systems — for safety against electrical discharges.
Sensors and cameras — in smart grid networks, for real-time monitoring.

2025
The
Peak
of
Data
and
AI
3. REPORT GENERATION (MANUAL)
Fill in an excel template file
4. GENERATE SHAPE FILE
Containing the information linked to the utility poles
5. ASSET BASE
Correction of the asset base – Data Base
6. USER NOTIFICATION
Notify user by email of results

2025
The
Peak
of
Data
and
AI
Solution
Data Integration/ETL + GeoAI
Picterra

2025
The
Peak
of
Data
and
AI
Picterra
From Earth observation imagery to actionable insights with AI
Founded in 2016 in Switzerland
• +100 corporate clients globally
• + 30.000 Machine Learning Models

2025
The
Peak
of
Data
and
AI
Products
Detect objects, patterns, and
changes faster than ever by
managing the entire geospatial ML
pipeline with our cloud-native
platform.
Analysis of plots related to
sustainability compliance, such as
EUDR and carbon storage estimates.
It can be integrated with various
software and platforms, mainly
because it uses open standards
(REST, JSON, GeoJSON, shapefile),
which allows for great flexibility.
GEOAI Tracer
API

2025
The
Peak
of
Data
and
AI
Integration Picterra and FME
Drone Images Build Detectors ML FME Output
Perform object detection with PicterraConnector within FME:
• List folders, rasters, detectors and vector layers.
• Upload images and detection áreas.
• Perform detection.
• Download results.

2025
The
Peak
of
Data
and
AI
Solution
Extract Images Identify Objects
Receving the
results
Email
3
4
5
1 2
Update Excel
file template

2025
The
Peak
of
Data
and
AI
Thank
You
Danilo de Lima & Thiago Brigagão
Solutial Soluções e Análise de Dados
danilo@solutial.com.br
thago@solutial.com.br

Connecting Data and Intelligence: The Role of FME in Machine Learning

More Related Content

Similar to Connecting Data and Intelligence: The Role of FME in Machine Learning (20)

More from Safe Software (20)

Recently uploaded (20)

Connecting Data and Intelligence: The Role of FME in Machine Learning