Hybrid ML + Image Processing Predicts Fruit Freshness with 90%+ Accuracy

Fruit spoilage causes massive economic losses in agriculture. To address this, researchers Amir Reza Hashemi and Shahram Amiri introduced a hybrid approach that fuses traditional image processing with deep learning. Their system first uses an image processing algorithm to quantify spoilage on a continuous 0 (fresh) to 100 (rotten) scale. Separately, a convolutional neural network (CNN) performs binary classification (fresh or rotten) on a large fruit image dataset. The outputs are combined via logistic regression to improve accuracy, and that regression model then enables the image processing algorithm alone to output binary decisions in real time—eliminating the need for the CNN during deployment.

Validated on apples and oranges, the system achieves over 90% accuracy while running in real time on modest hardware—no high-end GPUs required. The key current limitation is that fruits must be isolated on a white or transparent background, which the authors note could be addressed in future work with advanced segmentation models. This work demonstrates how simple image processing techniques paired with machine learning can deliver practical, low-cost solutions for agricultural quality monitoring, potentially reducing waste and improving supply chain efficiency.