Leaf Classification for Plant Recognition with Deep Transfer Learning

Plant recognition systems that developed by computer vision researchers, help botanists in faster recognition and detection of unknown plant species. Until now, multiple studies focused on the process or algorithms that maximize use of botanical datasets for plants prediction modeling, but this method depends on leaf characteristics which can be changed with botanical data and different feature extraction techniques. On the other hand, recently, due to the popularity and successful implementation of deep learning-based methods in various areas such as image classification, object detection, and speech recognition, the researchers directed from traditional feature-based methods to deep learning. In this research, one more efficient method presented that use transfer learning to recognize plant for leaf classification, which first uses a pre-trained deep neural network model for learning useful leaf characteristics directly from the input data representation. Then use a logistic regression classifier for leaf classification. It is seen that transfer learning from a large dataset to limited botanical dataset in plant recognition task is well done. The proposed method is evaluated on two well-known botanical datasets, i.e., Flavia with 32 classes and Leafsnap with 184 classes, and has succeeded in achieving an accuracy of 99.6% and 90.54%, respectively. The results show that despite the large change in the number of classes in these two datasets, the proposed method, have a good performance and show the better result than methods based on the hand-crafted feature and other methods based on the deep learning in terms of memory and precision.