Fine-Tuned MobileNet Classifier for Classification of Strawberry and Cherry Fruit Types

Abstract

This paper proposed an accurate, fast and reliable strawberry, cherry fruit detection and classification system for the automated strawberry cherry yield estimation. State-of-the-art deep learning-based fine-tuned MobileNet Convolutional Neural Network is developed to detect and classify strawberry and cherry fruit types in the outdoor field. The proposed CNN model is trained on 4250 strawberry fruit images, 3878 Cherry fruit images and tested on 990 strawberry fruit’s images, 1012 Cherry fruit images. To capture features and classify fruit type, a fine-tuned MobileNet Convolutional Neural Network model is presented in this study. The original MobileNet CNN model has 88 layers, which is computationally intensive and has more parameters. In the fine-tuned MobileNet CNN model, top layers are frozen and few layers are replaced with other layers such as a depthwise layer, pointwise layer, ReLu and Batch normalization layer, global average pooling layer. The fully connected layer is removed. The fine-tuned MobileNet CNN model performs quite well with higher accuracy of classification of fruit at less computation cost. The proposed CNN Model performs classification and labels them as Blueberry, Huckleberry, Mulberry, Rasberry, strawberry, strawberry wedge, Cherry Brown, Cherry Red, Cherry Rainier, Cherry wax Black, Cherry wax Red, Cherry wax Yellow. The proposed model’s average validation accuracy is about 98.60% and the loss rate is about 0.38%. The fruit images are acquired from the cultivation field include fruits that are occluded by foliage, under the shadow and some degree of overlap of strawberry, cherry flowers.