Computer Laboratory

Remote sensing platform with Raspberry-Pi Network: CO2


Sub-Saharan Africa depends heavily on agriculture, providing income to almost 80% of its population. Small holder farmers who grow under open irrigation constantly face pests and diseases that cause 50% loss of their yield on average. Greenhouses are therefore catching up as means to protect crops from pests and diseases. Farmers reduce crop losses by almost 20% but further improvements are limited, because farmers have to spot pests and diseases by eye to take the action. Farmers also try to prevent the use of chemicals in order to keep the crop organic. A majority of these pests are difficult to spot and may only become apparent once they have fully infested. BUGALERT project develops a smart greenhouse monitoring solution using image recognition to readily pick up pest and disease infestation and alert the farmers. The resulting plant images can also be used to advice on proper feeding patterns to ensure leaves remain green and healthy. The proposed system uses networked Raspberry Pi computers. Raspberry Pi is an inexpensive, tiny computer. We will install a camera and sensors to each Raspberry Pi, which takes pictures of crops and senses environmental data. Raspberry Pis can communicate with each other for data transfer. The collected data will be aggregated at the gateway-node in the greenhouse, where data can be partially analysed and sent to the main station by wireless communication. At the main station, the aggregated data will be analysed using advanced data analysing techniques such as machine learning. The whole operation will be integrated as a pest/disease alert system to proactively communicate with farmers.

Goal and Objectives

  1. Build a smart monitoring system for greenhouse farming:
    • Capture image of crops by cameras linked with Raspberry Pi.
    • Analyse image data to identify specific symptoms of pests/diseases.
    • Build efficient data-transfer mechanism within greenhouse and between greenhouse and data analysis site (and/or Cloud).
  2. Early alerting system on pest and disease detection to prevent extensive damage:
    • Exploit Machine Learning technique to detect/predict pests/diseases.
    • Refine image recognition neural network model for specific pests/diseases.
    • On-site inference on identifying pests/diseases.
    • Accumulation of sample image data to build a database which can be shared by various greenhouses.
  3. Deploy system to various parts of Kenya and analyse different data to extract beneficial information:
    • Obtain user-centred feedback on the system and information provided.
    • Access to small holder networks in Kenya through our established partnerships.
    • Transfer technology to users beyond greenhouses, including farmers using drip irrigation technology.

Cambridge Experiment in Computer Laboratory

Four Raspberry Pi sensing platform is deployed in Computer Laboratory.

System Overview

The experiment in Cambridge consists of 4 Sensor Raspberry Pis and 1 Gateway Raspberry Pi. The Sensor Raspberry Pis publish MQTT messages to the Gateway Raspberry Pi.

The Sensor Raspberry Pis collect various data, for example photos, soil moisture, co2 ppm etc. The sensor Pis can be customised through the use of lua scripts, in order to for example only take photos during sunlight hours.

The Gateway Raspberry Pi acts as an MQTT bridge to a Cloud Server. The gateway forwards the MQTT messages to the cloud, as well as storing the MessagePack payload locally on the Pis SD card

The Cloud Server takes the MQTT messages and parses the MessagePack payload data and stores it in MongoDB.

The images and database from the Cloud Server are periodically backed up to a server in Cambridge. This is done through the use of rsync. The Mongo database on this server is then populated with the latest version from the cloud.

Technical Specification

User Guide - contains how to issue commands to sensors, generate CSV files and alter website

Technical Specification