Deep Learning for Plant Classification and Content-Based Image Retrieval

摘要

The main goal of the present research is to classify images of plants tospecies with deep learning. We used convolutional neural network architectures forfeature learning and fully connected layers with logsoftmax output for classification.Pretrained models on ImageNet were used, and transfer learning was applied. In thecurrent research image sets published in the scope of the PlantCLEF 2015 challengewere used. The proposed system surpasses the results of all top competitors of thechallenge by 8% and 7% at observation and image levels, respectively. Oursecondary goal was to satisfy the users’ needs in content-based image retrieval togive relevant hits during species search task. We optimized the length of the returnedlists in order to maximize MAP (Mean Average Precision), which is critical to theperformance of image retrieval. Thus, we achieved more than 50% improvement ofMAP in the test set compared to the baseline.