نبذة مختصرة : Cotton mechanical topping is one of the most important cultural practices to improve crop yield during production. The shoots of cotton topping can be cut at about 10–20 cm from the top of plants. However, the performance of mechanical topping has been limited to computing power and real-time transport in several edge-moving devices at present. The detection can also be confined to the motion blur and small target occlusion. In this study, a lightweight detection model of a cotton bud (named CottonBud-YOLOv5s) was proposed using the well-known YOLOv5s architecture. Both performance and efficiency were optimized to detect the cotton buds in complex field environments. The ShuffleNetv2 backbone network was utilized to enhance the overall performance of the CottonBud-YOLOv5s model. The computational complexity was reduced to maintain the high accuracy of detection. In addition, the DySample dynamic upsampling module was integrated to replace the original ones. The computational costs were further reduced to improve the speed of detection. As such, the improved model was run more efficiently on edge devices with limited computing power. Real-time performance was also achieved during cotton mechanical topping. Moreover, the ASFFHead detection head and GC (global context) attention mechanism were also introduced into the head and neck components, in order to handle the varying object scales and complex contextual information. The scale invariance was significantly improved to extract the context-based features, which was crucial to detect the small targets that occluded or blurred due to the various motions in fields. Ultimately, the robustness of the model was improved to perform the best in real-world conditions. A series of ablation and comparison tests were conducted to validate the efficacy of the CottonBud-YOLOv5s model. The experimental results demonstrated that the introduction of the ASFFHead detection head and the GC global attention mechanism led to notable improvements in detection accuracy. Specifically, the average precision (AP) at 0.5:0.95 for small targets increased by 3.6 percentage points, while the average recall rate (AR) at the same threshold was improved by 2.1 percentage points. In the medium-sized targets, the AP and AR increased by 4.1 and 3.5 percentage points, respectively. In the large targets, the AP and AR increased by 6.5 and 5.9 percentage points, respectively. The improved model performed the best to detect the targets across a range of sizes. Furthermore, the CottonBud-YOLOv5s model shared significant improvements in the detection speed, compared with the state-of-the-art detection models, including Faster-RCNN, TOOD, RTDETR, YOLOv3s, YOLOv5s, YOLOv9s, and YOLOv10s. Specifically, the speed outperformed with the increases of 26.4, 26.7, 24.2, 24.8, 11.5, 18.6, and 15.6 frames per second, respectively. Additionally, the mean average precision (mAP) was improved by 14.0, 13.3, 5.5, 0.9, 0.8, 0.2, and 1.5 percentage points. The recall rate substantially increased by 16.8, 16.0, 3.2, 2.0, 0.8, 0.5, and 1.2 percentage points, respectively. Overall, the CottonBud-YOLOv5s model achieved a remarkable mean average precision (mAP) of 97.9%, a recall rate of 97.2%, and a CPU detection speed of 27.9 frames per second, indicating exceptional performance in both accuracy and speed. Visual analysis confirmed that the CottonBud-YOLOv5s model excelled in various detection scenarios, including the single-plant, multi-plant, motion blur, and small target occlusion conditions. Its superior performance in these areas highlighted its robustness and effectiveness in realworld agricultural environments, where such challenges were commonly encountered. In conclusion, the CottonBud-YOLOv5s model can offer a promising solution to the precise, real-time detection of cotton buds in densely planted environments, indicating high detection accuracy, enhanced robustness, and efficient computational performance. The finding can provide a solid visual detection for cotton mechanized topping in automated agricultural practices. [ABSTRACT FROM AUTHOR]
نبذة مختصرة : 针对棉花机械打顶作业过程中, 边缘移动设备算力受限实时性差, 运动模糊、小目标遮挡导致难以检测的问题, 该研究基于 YOLOv5s 模型提出 CottonBud-YOLOv5s 轻量型棉花顶芽检测模型, 该模型采用 ShuffleNetv2 主干网络和 DySample 动态上采样模块替换原始模块降低计算量, 提高模型检测速度; 头部 (head) 和颈部 (neck) 分别引入 ASFFHead 检测头和 GC (global context) 全局注意力模块增强模型尺度不变性和上下文特征提取能力, 提高小目标遮挡 和运动模糊图像的检测性能。通过消融试验和模型对比试验, 验证 CottonBud-YOLOv5s 棉花顶芽检测模型的可行性。 试验结果表明: 引入 ASFFHead 检测头和 GC 全局注意力机制后, 小目标平均精度 AP0.5:0.95 和平均召回率 AR0.5:0.95 值比 引入前分别提升 3.6、2.1 个百分点, 中目标平均精度 AP0.5:0.95 和平均召回率 AR0.5:0.95 值分别提升 4.1、3.5 个百分点, 大 目标平均精度 AP0.5:0.95 和平均召回率 AR0.5:0.95 值分别提升 6.5、5.9 个百分点; 与 Faster-RCNN、TOOD、RTDETR、 YOLOv3s、YOLOv5s、YOLOv9s 和 YOLOv10s 检测模型相比检测速度分别提升 26.4、26.7、24.2、24.8、11.5、18.6、 15.6 帧/s, 平均精度均值分别提升 14.0、13.3、5.5、0.9、0.8、0.2、1.5 个百分点, 召回率分别提升 16.8、16.0、3.2、2.0、 0.8、0.5、1.2 个百分点, CottonBud-YOLOv5s 模型平均精度均值达到 97.9%, 召回率达到 97.2%, CPU 检测速度达到 27.9 帧/s。由模型可视化分析可知 CottonBud-YOLOv5s 模型在单株、多株、运动模糊、小目标遮挡的整体检测性能优于 其他检测模型。该模型具有较高的检测精度、鲁棒性和检测速度, 适用于密植环境下棉花顶芽的精准检测, 可为棉花机 械化打顶提供视觉检测基础。 [ABSTRACT FROM AUTHOR]
Processing Request
No Comments.