مستشعر الأشباح

T.ONE Ltd

Contains ads

About this app

Environmental Analyzer is an advanced science app that integrates over 15 measurement sources from your phone into a single analysis engine, giving you a real-time understanding of what's happening around you—sound, magnetic field, motion, light, and radio energy.

How does it work?

1. Real-time Sensing — The app collects readings from your magnetic compass, accelerometer, gyroscope, barometric pressure, light, microphone, camera, Wi-Fi networks, Bluetooth devices, device temperature, and geolocation, every 100 milliseconds.

2. Smart Learning — During the first few minutes of calibration, the app builds a "baseline" for your natural environment: What's the typical decibel level in your room? What's the strength of the magnetic field on your street?

3. Anomaly Detection — After learning, it compares each new reading to the baseline, calculates an anomaly score for each source, and then combines them into a single index from 0 to 10. When the threshold is exceeded, it records the entire event with the time, location, snapshot, and values.

4. Documentation and Analysis — View events on the map, monitor live graphs for any sensor, and export a PDF, CSV, or JSON report for documentation.

Uses

- Documenting noise and auditory disturbances with verifiable figures
- Monitoring vibrations and movement while traveling or leaving your luggage
- Examining the work and home environment (lighting, pressure, temperature)
- Monitoring the magnetic field and detecting large metal objects
- Detecting unusual changes in your surroundings at night
- Research and educational projects for physics and engineering students

Features

- Live general environmental indicator with audible and vibration alerts
- Live graph for each individual sensor
- FFT sound spectral analysis with dominant frequency detection
- Visual analysis: motion, edges, distortion, optical flow
- Background operation with a customizable persistent notification
- Exporting professional reports and sharing with one click
- Full Arabic interface and RTL support

Scientific Integrity

The application only measures what the phone's hardware can actually measure. We are transparent about this: sound readings are estimated and not lab-calibrated, and some sensors may not be available on all devices (such as humidity). The application does not claim to see what the phone's camera cannot see.

Privacy First

Built on a real autoencoder neural network in Kotlin that runs and trains on the same machine during calibration, while preserving the TFLite path for any future external model. Checks how the classifier is called
A complete Kotlin Autoencoder network that trains on-device with TFLite support intact

AutoencoderNet Neural Network:

12 → 8 → 3 → 8 → 12 architecture — a bottleneck (3D) that forces it to absorb the "core of your natural environment"
Full manual backpropagation implementation, Xavier initialization, static seed
Each analysis frame = training sample (~10 samples/second)
Anomaly Detection Mechanism:

Reconstruction error = anomaly score (the network reconstructs "normal" correctly but fails with "abnormal").
The error is statistically normalized (Welford mean/deviation) and then logistic at 2σ → 0–10.
Ghost learning protection: High-error frames are trained at a 20x lower learning rate so they don't "get used to" the entity! A smart three-layer transition in AnomalyClassifier:

TFLite → Neural Network → Statistical Fusion
After 500 frames: Confidence begins to gradually transition to the neural network.
After 1500 frames: Full confidence is gained (statistical fusion remains a safety net).
Practically speaking: Start a chase → During the calibration (3 minutes), the network trains on your home environment → Afterward, any deviation from your unique environmental signature is detected more accurately and intelligently than static equations. I've also set up persist()/restore() to save the network's weights between sessions.

All processing and analysis happens entirely on your device. No calculations, no servers, no data transmission—public OpenStreetMap maps are used solely for displaying event locations.