Authors

Abstract

The core of intelligent and accurate plant protection of pests is the accurate identification of pest monitoring and early warning model, and the quality of pest sample image is crucial to the model identification accuracy. To solve the problem of complicated background and low contrast colour image samples, in this paper it is proposed a pest accurate segmentation method based on critical point nonlinear enhancement. The segmented image is used as the sample image of the Faster R-CNN model, which can improve the accuracy of the recognition model. Firstly, the original image is segmented by a strong classifier and the image of pest cells with calibrated grids is obtained. Secondly, the Spline adjustment curve is fitted according to the core gray scale range and critical point, and the contrast between pest and mesh in pest monomer image is enhanced based on the Spline adjustment curve. Finally, there are some operations for the enhanced image such as threshold segmentation, contour extraction, morphological transformation and others to obtain the pest image without background interference, and some segmentation experiments are performed to the pest image based on different segmentation methods. The experimental results show that the proposed method can accurately segment the pests in complex background, and the comprehensive evaluation indexes such as recall ratio and precision rate are greater than or equal to 91.5%, which is better than the traditional segmentation method.

Abstract in Chinese

虫害智能精准植保的核心是虫害监测预警模型的精准识别，而虫害样本图像的质量是决定模型识别精度的关键。为提高企业现有5G虫害监测装备的性能，在不对设备进行升级更新的情况下，解决其图像样本背景复杂、色彩对比度低等问题，获得无背景网格干扰的高质量的虫害样本图像，本文提出一种基于临界点非线性增强的虫害精准分割方法。将分割后的图像作为Faster R-CNN模型的样本图像，丰富样本集，提高识别模型精度。首先，使用强分类器对原始图像进行初步分割，获得含标定网格的虫害单体图像；其次，根据核心灰度范围与临界点拟合Spline调整曲线，将虫害单体图像基于Spline调整曲线增强虫体与网格的对比度。最后，对增强后的图像进行阈值分割、轮廓提取、形态学变换等操作，获得无背景干扰的虫害图像，并基于不同分割方法对虫害图像进行分割试验。试验结果表明：本文所提方法能在复杂背景中准确分割虫体，且查全率、查准率等综合评价指标均大于等于91.5%，分割效果优于传统分割方法。