Offline based on-field fruit ripeness detection system

This paper presents an offline-based on-field ripeness detection system using a convolutional neural network (CNN) machine learning algorithm and an ESP32 camera. The CNN model was developed by collecting datasets from various sources such as Kaggle and Chrome, and a multi-class CNN algorithm was created to classify three different fruits: apple, orange, and banana. The model achieved an overall accuracy of 92% across the eight different fruit classes. However, the size of the generated .h file was too large, which was not compatible with the memory of the ESP32, which necessitated narrowing the classification to just one fruit, in this case, apple. This resulted in near- perfect accuracy (close to 100%), and a reduced .h file size of 273kb, which was compatible with the ESP32.The camera captured images of apples and sent them to the ESP32 microcontroller for analysis using the trained CNN algorithm. The system achieved high accuracy in detecting the ripeness of apples in real-time, which can improve the efficiency of harvesting and reduce the wastage of produce. The system was programmed using the Arduino IDE, and it is cost-effective, and can be easily integrated into existing harvesting workflows. The proposed system has the potential to significantly improve the efficiency and sustainability of the agricultural industry.