Plant Leaf Classification through Deep Feature Fusion with Bidirectional Long Short-Term Memory

Abstract

Manual plant species classification requires more time, effort, and professional knowledge and also, and they are very costly. At present days, researchers utilized deep learning techniques for the classification of plants over plant images. The deep learning models attain great success, and then, the lack of interpretability set a limit on their application. To overcome these limitations, they utilized measurable, interpretable, and computer-aided features from plant leaf images. Image processing became very complex and crucial at the time of feature extraction. There are nearly 391,000 vascular plant species present in the world widely. So, the classification and identification of plants became complex and impractical for professionals. Most of the plant species have huge similarities and it consumes a huge amount of time for classification. So, it is essential to develop a computerized system for the identification and classification of plants. A great advancement is widely developed in science and technology for the recognition and classification of plant species in biological fields. Thus, the automatic plant leaf detection models are utilized to help professionals and botanists to classify plant species rapidly. This work proposes a novel automatic plant leaf image classification method through a deep learning algorithm. At first, the input images are gathered from the standard dataset, where the deep features from the gathered images are extracted using Resnet and VGG16. The extracted features are fused and fed to the classification stage. The plant leaf image classifications are done through the Bidirectional Long Short-Term Memory (Bi-LSTM). The empirical outcomes of the developed model have achieved better performance regarding precision and accuracy.