Semantic Segmentation of Plant Leaves Based on Generative Adversarial Network and Attention Mechanism

Abstract

Due to the rapid growth of the population, the pressure on food is increasing and the demand for higher crop yields is rising. It is crucial to periodically monitor plant phenotypic traits, and deep learning has a good effect on image recognition and segmentation. This paper proposes a method based on generative adversarial network and attention mechanism to improve the accuracy of semantic segmentation of plant leaves. First, the data set is standardized and divided into a training set and test set. The generator that produces the confrontation network uses Segnet as the backbone network and adds an attention mechanism to extract the phenotypic characteristics of plants. The discriminator utilizes a dual-input fully connected layer for true and false estimate. The experimental results show that compared with the original Segnet segmentation network, the proposed strategy improves the precision of pixel recognition PA. Also, the suggested technique has a high level of robustness and feature extraction precision. In addition to providing technical assistance for future crop cultivation and breeding, monitoring crop growth, ensuring yields as well as solving the food shortage problem.

Keywords

• Semantic segmentation
• attention mechanism
• generative adversarial network
• Segnet
• phenotypic characteristics