|
Abstract Maturity judgement is an important basis for mango harvesting and storage. In order to meet the limitation of hardware computing capacity of mobile devices, the performance of traditional machine learning and transfer learning in mango maturity identification was compared. The optimal model was selected and the mango maturity classification software was developed. One hundred images of Xiaotainong mango in different maturity were collected and then divided into training set and test set with a ratio of 8∶2 after data expansion. Accuracy, F1 value and prediction time were used as model evaluation indexes. Four machine learning algorithm models (K-NearestNeighbor, Support Vector Machine, Naive Bayes and Decision Tree) and five transfer learning algorithm models (AlexNet, ResNet18, VGG16, GoogleNet, and SqueezeNet) were used to train and test respectively. The performance of each model was compared and analyzed. The results showed that although machine learning had strong computing ability, the classification accuracy of machine learning was significantly lower than that of transfer learning. The classification accuracy of all transfer learning was >90%. Considering both classification accuracy and computing power, resnet18 performs best with accuracy rate reaches 98.75% and test time is only 74.66 ms after 20 iterations, which is superior to other transfer learning models.
|
|
|
|
|
|
2022, Vol. 35 
