QEFS: A novel plant disease prediction approach using quantum-inspired evolutionary feature selection

Plant disease prediction is crucial for global food security, prompting the development of novel detection techniques. Initially, Convolution Neural Networks (CNNs) were extensively employed in this domain for their image recognition and object detection capabilities. Recently, with the evolution of Quantum Computing (QC), Quantum Convolutional Neural Networks (QCNNs) have demonstrated improved classification and prediction performance in classical problems, such as medical image analysis and drug discovery. QCNNs excel in classification by leveraging effective features generated through their layers. Features extracted from QCNNs utilize quantum parallelism to explore various feature combinations simultaneously, enhancing the network’s ability to capture intricate patterns and relationships in classical image datasets. However, the high-dimensional nature of these quantum-derived features necessitates effective Feature Selection (FS) to address the curse of dimensionality, improve model interpretability, and ensure computational efficiency in downstream tasks. The study presents Quantum-Inspired Evolutionary Feature Selection (QEFS), a unique method combining effective quantum feature extraction along with the FS approach employing an evolutionary algorithm to tackle this challenge. A hybrid evolutionary optimizer is formulated by integrating key attributes from fundamental optimizing algorithms, combining the strengths of basic optimizers to yield an enhanced hybrid algorithm. In this methodology, features are initially extracted using a QCNN model. Subsequently, these features undergo an FS process using the hybrid FS approach to determine the optimal feature count. The selected features are then fed into five Machine Learning (ML) classifiers for classification. To validate the effectiveness of this approach, the study leverages two distinct plant datasets—normal plants and medicinal plants. The primary objective of the research is binary as well as multi-class classification, specifically differentiating between healthy and diseased plant images. This methodological innovation aims to overcome current limitations in QCNN applications through effective FS using the evolutionary optimization technique.