Narrow Results

The Chenopodiaceae species are ecologically and financially important, and play a significant role in biodiversity around the world. Biodiversity protection is critical for the survival and sustainability of each ecosystem and since plant species recognition in their natural habitats is the first process in plant diversity protection, an automatic species classification in the wild would greatly help the species analysis and consequently biodiversity protection on earth. Computer vision approaches can be used for automatic species analysis. Modern computer vision approaches are based on deep learning techniques. A standard dataset is essential in order to perform a deep learning algorithm. Hence, the main goal of this research is to provide a standard dataset of Chenopodiaceae images. This dataset is called ACHENY and contains 27030 images of 30 Chenopodiaceae species in their natural habitats. The other goal of this study is to investigate the applicability of ACHENY dataset by using deep learning models. Therefore, two novel deep learning models based on ACHENY dataset are introduced: First, a lightweight deep model which is trained from scratch and is designed innovatively to be agile and fast. Second, a model based on the EfficientNet-B1 architecture, which is pre-trained on ImageNet and is fine-tuned on ACHENY.

The experimental results show that the two proposed models can do Chenopodiaceae fine-grained species recognition with promising accuracy. To evaluate our models, their performance was compared with the well-known VGG-16 model after fine-tuning it on ACHENY. Both VGG-16 and our first model achieved about 80% accuracy while the size of VGG-16 is about 16$\times$ larger than the first model. Our second model has an accuracy of about 90% and outperforms the other models where its number of parameters is 5$\times$ than the first model but it is still about one-third of the VGG-16 parameters.

A large training sample is prerequisite for the successful training of each deep learning model for image classification. Collecting a large dataset is time-consuming and costly, especially for plants. When a large dataset is not available, the challenge is how to use a small or medium size dataset to train a deep model optimally. To overcome this challenge, a novel model is proposed to use the available small size plant dataset efficiently. This model focuses on data augmentation and aims to improve the learning accuracy by oversampling the dataset through representative image patches. To extract the relevant patches, ORB key points are detected in the training images and then image patches are extracted using an innovative algorithm. The extracted ORB image patches are used for dataset augmentation to avoid overfitting during the training phase. The proposed model is implemented using convolutional neural layers, where its structure is based on ResNet architecture. The proposed model is evaluated on a challenging ACHENY dataset. ACHENY is a Chenopodiaceae plant dataset, comprising 27030 images from 30 classes. The experimental results show that the patch-based strategy outperforms the classification accuracy achieved by traditional deep models by 9%.