Accurately Identifying Sound vs. Rotten Cranberries Using Convolutional Neural Network

Abstract

Cranberries, native to North America, are known for their nutritional value and human health benefits. One hurdle to commercial production is losses due to fruit rot. Cranberry fruit rot results from a complex of more than ten filamentous fungi, challenging breeding for resistance. Nonetheless, our collaborative breeding program has fruit rot resistance as a significant target. This program currently relies heavily on manual sorting of sound vs. rotten cranberries. This process is labor-intensive and time-consuming, prompting the need for an automated classification (sound vs. rotten) system. Although many studies have focused on classifying different fruits and vegetables, no such approach has been developed for cranberries yet, partly because datasets are lacking for conducting the necessary image analyses. This research addresses this gap by introducing a novel image dataset comprising sound and rotten cranberries to facilitate computational analysis. In addition, we developed CARP (Cranberry Assessment for Rot Prediction), a convolutional neural network (CNN)-based model to distinguish sound cranberries from rotten ones. With an accuracy of 97.4%, a sensitivity of 97.2%, and a specificity of 97.2% on the training dataset and 94.8%, 95.4%, and 92.7% on the independent dataset, respectively, our proposed CNN model shows its effectiveness in accurately differentiating between sound and rotten cranberries.

Keywords

• agricultural image analysis
• convolutional neural network
• fruit rot classification
• hyperspectral imaging
• machine learning
• OpenCV