Multilevel detection and classification of diseased plant leaf images using high-resolution superlet transform and E-ResNet

Plant leaf disease detection in early stage is a very challenging task that and is one of the most promising research areas in the field of precision agriculture. Advancements in the field of artificial intelligence improves the practice of precise plant protection and growth. Different deep-learning architectures are proposed in the recent past to detect and classify the diseased plant images. These architectures cannot detect multiple occurrences of a particular disease in an image. Also, these methods suffer from different illumination conditions, a complex background, and location of the disease. To overcome these limitations, this paper proposed a novel framework to detect and classify diseased plant images effectively. The proposed framework incorporates a combination of LBP, HOG, and GLCM features which can easily locate the diseased part from the image even in the complex background. Further, the framework utilizes superlet transform (SLT) which provides an enhancement in the classification accuracy by generating high resolution time–frequency spectrograms of the fused features. Further, these spectrograms help in detection of multiple occurrences and easily classify the diseased plant images using Enhanced-ResNet (E-ResNet) network. To validate the performance of the proposed framework experiments are conducted on PlantDoc dataset which provided an accuracy of 97.4% and 98.1% with VGGNet and E-ResNet classification model respectively. To further validate the efficiency and efficacy of the proposed framework an ablation study is also carried out.