HYDRO·CV
Space Agriculture · Environmental Engineering

Autonomous Hydroponic Plant Growth System

A closed-loop growth chamber that senses its own climate, recirculates nutrients, and measures plant biomass with a camera. No harvest needed to know how much the plant weighs. Built for long-term space travel.

Drag to orbit · tap a part
6.9%Biomass MAPE
0.91R² canopy→mass
6Independent trials
72hClimate log
21dGrowth cycle
4Sensor channels
Electrical integration

One Teensy runs the whole chamber

The BME680 talks to the Teensy 4 over a shared I²C bus at 3.3 V. The same controller switches the LED through a relay and drives the air-circulation fan. Click any wire or block.

Top-down camerafixed geometry image data Teensy 4 central control logic data acquisition relay switching 3.3VSCL 19SDA 18GND BME680 temp · RH eCO₂ · VOC VINSCLSDAGND I²C · 3.3 V logic · shared bus Relayswitches LED Blurple LEDconstant spectrum 5 V fansimulated breeze Submersible pumpcontinuous recirculation

Wire inspector tap to select

Computer-vision pipeline

From a green blob to grams

A fixed top-down camera and a constant blurple LED remove geometry and lighting noise. Every step below runs live in your browser on a generated plant image. Drag the day slider and watch the canopy area and predicted mass follow.

Day 12
Canopy0px
Area0cm²
Biomass0g

ExG = 2G − R − B
Area(cm²) = pixels × 0.0125
Biomass(g) = 0.14 · Area + 0.6

The pipeline steps and formulas are from the project. The calibration factor and regression coefficients here are illustrative stand-ins, since the fitted values are not published on the board.

Predicted growth curve illustrative

Results & validation

Stable climate, accurate predictions

Six independent growth cycles, one model, a mean error of 6.9%. The chamber held temperature and humidity steady enough for the camera to trust its own numbers.

Live sensor feed simulated stream

Values drift inside the ranges recorded on the chamber: 24–32 °C, 40–52 % RH, 400–750 ppm eCO₂, 30–230 ppb TVOC.

Climate stability 72 h monitoring

Temperature ±~3 °CHumidity ±~6 % RH

Curves are reconstructed to match the recorded ranges and variance.

Air quality dynamics 72 h monitoring

TVOC 30–230 ppb, non-accumulativeeCO₂ 400–750 ppm, transient peaks

Curves are reconstructed to match the recorded ranges.

Prediction error by trial MAPE %

Trial error (from the board)15% target

Predicted vs measured biomass R² = 0.91

Six points, one per trial. Errors match the board's per-trial MAPE. The gram values are illustrative because per-trial masses are not printed on the board.

Design targets vs results

Engineering decision analysis

Why these parts

Each choice went through a weighted decision matrix. Criteria weights are shown in the header row. Totals are the board's own scores.

Analysis & interpretation

What it proves, and where it stops

Engineering significance

  • Non-destructive biomass estimation in controlled environments
  • No destructive harvest needed during growth monitoring
  • A base for autonomous agricultural systems
  • Embedded systems, computer vision and predictive modeling in one framework

Limitations

  • Assumes constant lighting and fixed camera geometry
  • Leaf overlap adds segmentation uncertainty
  • Biomass depends on leaf density and water content
  • Small dataset, n = 6 trials
  • Error sources: edge segmentation, calibration sensitivity, small environmental swings, VOC/CO₂ sensor noise

Where it can go

  • Space-based agriculture (CELSS systems)
  • Autonomous greenhouse optimization
  • Precision agriculture monitoring
  • Resource-constrained environments