Transfer Learning-based Fruit Freshness Monitoring for Future Autonomous Industrial Robotic Arms

Abstract

Consumption of fruits is a daily routine in humans’ life. The freshness of any fruit impacts their consumers, which will spoil health. So, the manual checking of each fruit freshness in farm field is difficult task. So, future autonomous industrial robotic arms automatically check the freshness of fruits in farm fields and distinguish the normal, abnormal fruits using artificial intelligence (AD properties. However, the conventional AI models like machine learning (ML), deep learning (DL) were failed to classify the multiple fruit freshness classes due to improper feature analysis. So, this work is focused on development of transfer learning (TL) based fruit freshness detection (TL-FFD) with multiple classes in automated environment. Initially, fruit freshness dataset is considered with green, ripe, and over ripe classes. Then, dataset pre-processing operation is implemented, which normalizes all images with uniform size. Later, the TL based GoogleNet model is adopted to train the pre-processed dataset, which stores the memory of multiple classes. Then, the random test image is applied to check the validity of the system, where GoogleNet model predicts the class of test images. Here, the proposed TL-FFD method achieved the 98.82% of accuracy and the simulation results shows that the proposed TL-FFD outperformed as compared to other ML, and DL models.