Using Deep Learning-Based Features and Image Augmentation to Predict Brix Values of Strawberries for Quality Control

The fields of Computer Vision and Artificial Intelligence are rapidly developing technologies that hold promise for
enhancing the economic viability of the Agriculture industry. This is an initiative to help strawberry exporters and growers to
choose high-quality strawberries concerning their sweetness by automatically predicting the Brix values from their images.
Using a novel dataset of 150 Strawberry images and their corresponding Brix values as labels, a deep learning algorithm called
ResNet101 is utilized for feature extraction and different machine learning-based regression models are used for predicting Brix
values. The image dataset is generated using a Logitech C920 HD camera, and each sample's instrumental Brix readings are
collected using a Brix refractometer. Image augmentation is employed for the dataset enhancement. 70% of the entire dataset
is used for training, and the remaining 30% for testing. With a high degree of Brix prediction accuracy, 96.3142%, the squared
exponential GPR model is proven to be the best-fit model for this dataset. This method can significantly help provide highquality control requirements for the strawberry sector. An RMSE value of 0.4772 and a coefficient of determination value of 0.8648 are the obtained performance evaluation metrics values during the prediction phase. Also, the MAE and MSE values obtained are 0.0233 and 0.2277, respectively. These findings show the possibility of combining deep learning with image enhancement to increase the precision of Brix value predictions for strawberries, and they may be a useful tool for enhancing the effectiveness and precision of quality control measures in the fruit business