Smart Compost Monitor using Edge AI

Good compost is a managed biological process: the pile must pass through mesophilic, thermophilic and curing phases, stay at 50–60 % moisture, and be turned before the core goes anaerobic and starts producing methane and odour. Most home and campus composting fails because nobody measures any of this — the pile is either too dry, too wet, or never turned. This project instruments the pile: a weatherproof ESP32-S3 node with a temperature probe, moisture sensor, methane and oxygen sensors logs the pile's vital signs around the clock, and a TinyML image classifier on the same chip identifies each bucket of waste as it is added (greens vs browns), keeping the carbon-nitrogen balance visible. The dashboard shows the 21-day temperature curve with phase bands, flags when a turn is due, and computes a maturity index from temperature drop, pH and EC trends. The student gets a real embedded-ML project — data collection, model training, quantization, on-device inference — wrapped around a genuine environmental application. Because every pile, climate and feedstock differs, the kit makes no claim about composting speed or classifier accuracy out of the box: the buyer trains and validates the model on their own waste photos and reports their own metrics.

1. The sensor node is installed beside the compost pile with the temperature probe in the core, moisture sensor mid-pile and gas sensors in a vented sampling head.
2. Each time waste is added, the student photographs it with the node's camera; the on-device classifier labels the input and logs the greens/browns balance.
3. Every 10 minutes the node samples all sensors, runs the maturity and phase logic, and pushes readings to the dashboard over Wi-Fi.
4. The dashboard renders the 21-day temperature curve with phase bands, moisture trend with turning-event markers, and the current phase (mesophilic / thermophilic / cooling).
5. When thresholds are crossed, the dashboard raises a turn advisory; the student turns the pile, logs the event, and watches the temperature respond.
6. For the report, the student exports the logs, and trains the classifier in the provided notebook on their own labeled photos, reporting their own validation accuracy.

Tech stack:

• ESP32-S3 (TinyML inference + Wi-Fi) with camera module
• DS18B20 pile temperature probe, capacitive soil-moisture sensor
• MQ-4 methane and MQ-135 air-quality/gas sensors, oxygen sensor module
• Weatherproof enclosure with cable glands and probe leads
• TinyML image classifier (transfer-learned CNN, int8 quantized, ~48 KB)
• CompostSense web dashboard (phase tracking, maturity gauge, CSV export)
• pH and EC pocket meters for maturity measurements

• [On-device waste classifier] A quantized convolutional model on the ESP32-S3 classifies each waste input (vegetable greens, fruit peels, dry leaves, cardboard and more) with no cloud, no latency and no privacy concerns.
• [Pile vital-sign sensing] DS18B20 temperature probe, capacitive moisture sensor, MQ-series methane sensor and oxygen sensor capture the four variables that decide whether a pile thrives or stinks.
• [Composting-phase tracking] The dashboard plots the temperature curve against mesophilic, thermophilic and cooling phase bands, so the student can see the biology happening.
• [Turn recommendations] When core temperature climbs toward the high-60s or oxygen sags, the dashboard advises a turn within 24 hours — the single most useful composting intervention.
• [Maturity index] Temperature drop from peak, pH and EC readings combine into a 0–100 % maturity gauge with an estimated ready-date, giving the results chapter a quantitative endpoint.
• [Training notebook] A documented Python notebook covers dataset labeling, transfer learning, quantization to int8 and flashing the model — the full TinyML pipeline the student defends in the viva.
• [Weatherproof field node] IP-rated enclosure, cable glands and probe leads let the station live beside an actual compost pile or bin through the semester.

• Working smart compost monitor (ESP32-S3 node, camera, all sensors and probes, weatherproof enclosure)
• pH and EC pocket meters for maturity measurements
• TinyML training notebook (labeling guide, transfer learning, quantization, deployment)
• CompostSense dashboard source code (single-file web app, documented)
• Project report PDF (embedded-ML pipeline, compost science, buyer's validation method)
• PPT presentation for final review
• Viva Q&A preparation document (TinyML, quantization, compost phases, C/N ratio, sensor principles)
• Field installation and pile-building guide