基于改进的Mask R-CNN自然场景下苹果识别研究

Research on Natural Scene Apple Recongnition Based on Improved Mask R-CNN

在复杂自然场景下,苹果目标因具有成簇生长、重叠果实和光线变化大等特点,应用深度学习方法相比传统方法来实现果实的识别优势明显。提出基于Mask R-CNN网络检测分割架构,采用膨胀卷积的优化策略,通过候选框与像素分割相结合的思路,同时对输入苹果图像进行目标果实的识别。实验结果表明,基于Mask R-CNN框架改进的网络模型的识别性能较原始Mask R-CNN网络有较大提升。针对不同光照角度、不同颜色和不同大小的苹果,改进Mask R-CNN网络的F_(1)值分别提升了2.17%,1.87%和4.93%。

In the field environment,fruit images are easily affected by many external environmental factors such as light changes,fruit size difference,complicated background noise,the application of deep learning method has obvious advantages over traditional methods to realize fruit recognition.To address these problems,this paper proposes a detection and recognition framework based on Mask R-CNN network,which uses the dilated convolution optimization strategy and combines bounding box and pixel segmentation,simultaneously recognizes fruits from the input apple image.The experimental results show that the recognition performance of the improved network model based on Mask R-CNN framework is better than that of the original Mask R-CNN network.The F_(1) score of improved Mask R-CNN network was improved by 2.17%,1.87%and 4.93%for apples of different illumination angles,colors and sizes,respectively.