Enhanced Plant Image Recognition Using AlexNet and Deformable Convolution with Preprocessing Techniques

Abstract

Plant image recognition plays a vital role in botany and machine learning, yet challenges such as noise, occlusion, and unclear images limit classification accuracy. This paper proposes an improved deep learning framework based on the AlexNet convolutional neural network (CNN) for plant image recognition. To address issues of incomplete features and noise, we integrate deformable convolution layers and apply image preprocessing techniques, including Gaussian filtering and edge enhancement, using OpenCV and PyQt5. Transfer learning is employed to update the model with large-scale plant datasets, enhancing adaptability to diverse plant species. Experimental results demonstrate that the enhanced AlexNet model achieves over 98% accuracy in recognizing ten plant species, surpassing traditional CNN approaches in both efficiency and precision. The findings underscore the potential of combining advanced CNN architectures with preprocessing to achieve robust plant image classification.