Fruit ripeness detection and automated harvesting represent essential components of precision agriculture, yet manual methods remain labor-intensive, inconsistent, and inefficient. This paper introduces a smart agricultural solution that leverages edge computing and computer vision to automate ripeness detection and harvesting. The system employs an ESP32-CAM module to capture fruit images and performs real-time classification using HSV color space analysis. Once ripeness is detected, the data is transmitted via Wi-Fi to a NodeMCU microcontroller, which activates a robotic arm with a servo motor and soft gripper for precise harvesting. Ripeness data and harvesting events are logged in a centralized database, ensuring traceability and enhancing transparency. The system was evaluated using apples and mangoes, demonstrating high accuracy and adaptability across fruit types. By minimizing latency and reducing reliance on manual labor, the platform increases efficiency and consistency in the harvesting process. The integration of edge computing and automated actuation makes the system scalable, cost-effective, and well-suited for smart farming applications, advancing sustainability and productivity in modern agriculture.
Introduction
Overview:
Fruit ripeness detection is a major challenge in agriculture due to labor shortages and reliance on manual inspection. This project proposes a low-cost, lightweight solution using MobileNet CNN and edge computing to classify fruit ripeness and automate harvesting with a robotic arm. The system is built for real-time performance on resource-constrained devices like the ESP32-CAM.
Literature Survey Highlights:
Color-Based Detection: Early models used RGB and HSV color spaces but suffered from lighting sensitivity and processing delays.
Shape & Texture Analysis: More accurate but computationally heavy and sensor-dependent.
Deep Learning Models: CNNs improve accuracy but demand high resources, limiting real-time edge use.
Edge Computing Systems: ESP32-CAM-based solutions reduce latency and cost but often lack full automation.
Robotic Harvesting: Systems with robotic arms show promise but struggle with environmental variability and cost.
Hybrid Approaches: Combine machine vision, IoT, and robotics but face challenges with adaptability, cost, and complexity.
Proposed System:
A compact, efficient fruit harvesting system that:
Uses ESP32-CAM for real-time image capture and local HSV-based ripeness detection.
Employs Wi-Fi communication to send ripeness/location data to NodeMCU.
Controls a servo motor-driven robotic arm with a soft gripper for non-destructive harvesting.
Operates entirely on edge devices, reducing reliance on cloud infrastructure.
Modular design supports various fruit types and is well-suited for small-scale and remote farming.
Methodology:
Image Processing: Local image analysis using OpenCV in HSV space. Steps include resizing, noise filtering, thresholding, and contour detection.
Edge Computing: ESP32-CAM handles all processing locally; NodeMCU controls the robotic arm based on transmitted data.
Robotic Arm Control: Precision movement using servo motors guided by real-time feedback.
Plucking Mechanism: Soft gripper picks fruits gently and accurately, reducing waste and damage.
Implementation Details:
Dataset: 10,000 labeled fruit images (apples, mangoes) under diverse conditions.
Model Training: MobileNet with preprocessing (resizing, normalization, Gaussian filtering) and augmentation for robustness.
Hardware Components:
NodeMCU (ESP8266): Microcontroller for control logic and communication.
ESP32-CAM: Captures and processes images; supports real-time inference.
LCD Display: Shows ripeness and system data.
Power System: Includes transformer and rectifier for stable DC supply.
Servo Motor: Controls robotic arm with high torque and precise positioning.
Conclusion
In the realm of real-time fruit harvesting systems, recent advancements have redefined the possibilities of agricultural automation, driving the transition toward smarter and more efficient farming practices. By harnessing cutting-edge technologies such as computer vision, robotics, and edge computing, these systems have demonstrated remarkable capabilities in detecting fruit ripeness and performing precise harvesting operations. This progress represents a transformative leap in modern agriculture, reducing reliance on manual labor and ensuring consistent harvesting quality. However, amidst the celebration of these achievements, it is essential to address the persistent challenges in scalability, adaptability to diverse crop types, and the integration of sustainable practices, which remain critical focal points for future research and development.
References
[1] Wei Ji, Tong Zhang, Bo Xu, Guozhi He (2024). Apple recognition and picking sequence planning for harvesting robot in a complex environment
[2] Kaixiang Zhang, Kyle Lammers2, Pengyu Chu2, Zhaojian Li1, Renfu Lu An automated apple harvesting robot—From system design to field evaluation (2023)
[3] Takeshi Yoshida , Takuya Kawahara and Takanori Fukao Fruit recognition method for a harvesting robot with RGB-D camera(2022)
[4] HanwenKang ,Hongyu Zhou, and Chao Chen Visual Perception and Modeling for Autonomous Apple Harvesting(2020)
[5] Bu L., Chen C., Hu G. 2022. Design and evaluation of a roboticapple harvester using optimized picking patterns. Comput.Electron.
[6] Gangammanavar H., Sen S. 2021.Stochastic dynamic linear pro-gramming: A sequential sampling algorithm for multistage stochastic linear programming