Fruit Disease Classification using Lévy Flight Strategy based Osprey Optimization algorithm

An automated fruit identification system is extremely beneficial for supermarket consumers which helps to identify and quality of fruits. Recently, Deep learning models have significantly contributed to disease classification and pattern recognition. However, the existing models still suffer from the gradient vanishing problem, leading to classification performance biases. In this paper, the Levy Flight Strategy based Osprey Optimization Approach (LFOOA) is proposed to classify the disease from the fruits. The model was successfully trained and tested on the Fruit-360 dataset collected from the Kaggle database. The proposed approach enhances classification performance and minimizes bias by determining the optimal parameters. The obtained data are pre-processed using normalization which scales the features in an appropriate range. Then, the MobileNet and EfficientNet are used to extract the features and lastly, the CNN classifier is used to classify the fruit disease accurately. The proposed LFOOA achieves a high accuracy of 99.93 %, F1-score of 99.00, Recall of 99.15% and precision of 98.87% compared to the other existing models.