Authors

Abstract

The recognition of Agaricus bisporus is a key step in the intelligent picking of Agaricus bisporus. Given the complex background and limited computing resources of edge devices in actual planting scenarios, an improved Mask-RCNN model for Agaricus bisporus recognition was proposed. In this method, the backbone feature extraction network of the baseline Mask-RCNN model was replaced with the lightweight MobileNetV3 network to reduce the model complexity. Meanwhile, the BiFPN network was used to replace the original FPN feature fusion network, thereby strengthening feature fusion and enhancing the model's ability to learn image features and acquire contextual information. Experimental results showed that the improved Mask-RCNN model’s parameters and floating-point operations were 24.46M and 173G, respectively, which were 44.4% and 24.5% lower than those of the baseline Mask-RCNN model, and the frame rate increased by 3.55 FPS, indicating a better prospect for deployment on edge devices. This method can provide technical support for the development of the visual system of Agaricus bisporus picking robots.

Abstract in English

双孢菇识别是双孢菇智能采摘的关键环节。针对实际种植场景中背景复杂以及边缘设备计算资源有限的问题，提出了一种用于双孢菇识别的改进 Mask-RCNN 模型。该方法将基准 Mask-RCNN 模型的骨干特征提取网络替换为轻量化的 MobileNetV3 网络，以降低模型复杂度；同时，采用 BiFPN 网络替代原有的 FPN 特征融合网络，从而强化特征融合，提升模型对图像特征的学习能力和上下文信息的获取能力。实验结果表明，改进后的 Mask-RCNN 模型参数为 24.46M，浮点运算量为173G，分别较基准 Mask-RCNN 模型降低了 44.4% 和 24.5%，帧率提升了3.55 FPS，在边缘设备上具有更好的部署前景。该方法可为双孢菇采摘机器人视觉系统的研发提供技术支持。