IoT-Driven Smart Farming System to Transform Agriculture

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jan 10 2024

In a paper published in the journal Scientific Reports, researchers delved into the transformative role of data-driven methodologies in agriculture, countering prevailing assumptions. They detailed an innovative integrated approach harnessing the Internet of Things (IoT) equipped sensors, driving an intelligent agricultural management system tailored for predicting rainfall patterns and monitoring fruit health.

The proposed system leveraged AI models, notably employing a convolutional neural network (CNN) with a long short-term memory (LSTM) layer for rainfall prediction. Additionally, a CNN incorporating a SoftMax layer, alongside several deep learning pre-trained models, was employed for comprehensive fruit health monitoring. Notably, a combined model—utilizing a CNN + LSTM configuration and a multi-head self-attention mechanism—served as a dual-purpose predictor for rainfall and a recognizer for fruit health, proving highly effective.

Related Work

Past studies have explored various technologies to optimize water usage and refine precision in agriculture, transitioning from passive observation to active implementation and management. One approach involved a wireless sensor network employing temperature and soil moisture sensors managed through Zigbee protocol for automated drip irrigation.

Another approach focused on designing a computerized watering system by monitoring ground moisture while sensors tracked plant temperature and moisture levels for water content assessment. An IoT-based intelligent irrigation system enabled soil moisture regulation and remote monitoring via an online platform. Other studies delved into diverse sensor applications, low-cost indoor farm management, and rainfall estimation using multiple linear regression (MLR).

Integrated Agricultural Technology Framework

The methodology adopted for designing an advanced agricultural system integrating IoT, real-time databases, Android applications, rainfall prediction, fruit health monitoring, and a combined CNN + LSTM + Attention Layer system is detailed. The integrated IoT system comprises sensors such as soil moisture, temperature, humidity, and air quality alongside hardware components like NodeMCU, light-emitting diodes (LEDs), water pumps, and motor drivers. These sensors, embedded in different field areas, collect real-time environmental data, aiding an AI-based automation strategy. Researchers developed an Android app that enables real-time data monitoring and irrigation control. This app interfaces with a Firebase database for transmitting data and controlling pump status.

Researchers established a real-time database on the Google Firebase platform to synchronize and store data in a JavaScript object notation (JSON) format. This setup facilitates accessibility and data retrieval for the IoT system. The Android app interfaces with the Firebase database, allowing users to monitor data and control irrigation remotely.

The development process utilizes the Massachusetts Institute of Technology (MIT) App Inventor for initial app creation and design. It involves seamless integration with the Firebase platform, facilitating data access and management within the app. Throughout this process, the focus remains on crafting a user-friendly interface to enhance the overall experience, ensuring ease of navigation and interaction for the app's users.

Researchers approached rainfall prediction using machine learning techniques on weather datasets. Data preprocessing involves handling missing values and outliers, followed by model training and comparison using various classifiers like categorical boosting (CatBoost), gaussian naive Bayes (NB), random forest, and a CNN + LSTM network. A user-friendly web interface allows users to input parameters for rainfall prediction.

Using CNN models, researchers devised a system to differentiate fresh and rotten fruits for fruit health recognition. The dataset includes various images of bananas, apples, and oranges, each with distinct variations in shape and color. Preprocessing involves normalization, rescaling, and data augmentation, enhancing the dataset's diversity. TensorFlow is employed to build and train the fruit health recognition model, offering a web-based interface for users to assess fruit conditions.

Furthermore, a full-stack web development approach is adopted, creating three websites: Agricultural Management System, Rainfall Predictor, and Fruit Recognition. These websites provide comprehensive services, offering weather forecasting, fruit health recognition, real-time data monitoring, and irrigation control. The architecture incorporates hypertext markup language (HTML), cascading style sheets (CSS), JavaScript, and Flask for front-end and back-end development.

Finally, researchers designed a combined CNN + LSTM + Attention layer system, integrating sensor data and image inputs for rainfall prediction and fruit health recognition. The architecture harnesses the strengths of CNNs in pattern recognition, LSTMs in handling sequential data, and self-attention mechanisms for focusing on critical parts of input sequences, enhancing discrimination, and capturing long-range relationships.

IoT Performance and Impact

The IoT setup undergoes meticulous scrutiny in assessing the system's performance, validating its functionality and data monitoring capabilities. Monitoring acetone levels alongside vital sensor data empowers farmers with critical insights, while motor control guided by sensor readings enhances irrigation efficiency and logic for decision-making.

Evaluation of rainfall prediction models—CatBoost, Random Forest, and Gaussian NB classifiers—reveals CatBoost and CNN + LSTM models as top performers, guiding the choice for predicting rainfall events. Fruit health recognition models also demonstrate high accuracy in discerning fresh and rotten fruits.

Web interfaces like the Agricultural Management System and Fruit Recognition are comprehensive platforms offering weather forecasts, fruit health analysis, and real-time data access. Comparisons with previous work highlight the innovative aspects of this system, particularly the combination of CNN with LSTM and MHSA layers, elevating its precision and capabilities in predicting rainfall and monitoring fruit health.

The NodeMCU-based network has advantages in enhancing agricultural productivity and minimizing water waste. Its pivotal role lies in efficiently managing irrigation while aiding farmers in making informed decisions. The system highlights cost-effectiveness and provides reliable decision-making tools for farmers. It is essential to note its reliance on network coverage and hardware performance.

Conclusion

To summarize, the design entails a precision agriculture system that utilizes sensors, IoT, and ML algorithms to optimize crop yield through water regulation and process management. Integrating various sensors into an IoT framework allows real-time monitoring of an agriculture management system, focusing on rainfall prediction and fruit health monitoring.

The emphasis lies in designing and training ML algorithms—CatBoost, Gaussian NB, Random Forest, CNN + LSTM—for accurate prediction and tracking. Additionally, a combined rainfall predictor and fruit health recognizer employing CNN + LSTM with multi-head self-attention requires less memory and computation time, forming a robust integrated agriculture assistant accessible via a user-friendly web portal.