Abstract: This paper presents an enhanced real-time multilabel leaf classification system based on the Binary Relevance (BR) approach, efficiently implemented on the Xilinx PYNQ-Z2 FPGA platform. The multilabel classification task is decomposed into independent binary Support Vector Machine (SVM) models, each dedicated to identifying a specific leaf type, enabling modular scalability and parallel inference potential. Image preprocessing and feature extraction are performed using Python and OpenCV on the Processing System (PS), ensuring robust handling of diverse imaging conditions. Classification control logic is designed in Verilog and rigorously validated through simulation, guaranteeing precise sequencing and reliable hardware-software coordination. A user-friendly button-triggered prediction mechanism initiates on-demand classification, with results visually conveyed via onboard LEDs. Evaluated on both standard and custom datasets, the system demonstrates superior robustness against challenges such as low-light conditions, image rotation, and scale variations. Achieving high classification accuracy (>93%), ultra-low latency, and minimal power consumption, the proposed FPGA-based solution establishes a highly effective, deployable framework for embedded plant monitoring and precision agriculture applications.

Keywords: Binary Relevance, FPGA, Leaf Classification, PYNQ-Z2, Support Vector Machine, Embedded System, Real-time Prediction

Downloads: | DOI: 10.17148/IJIREEICE.2025.131220