YOLOv7-MA: Improved YOLOv7-Based Wheat Head Detection and Counting

Abstract

Detection and counting of wheat heads are crucial for wheat yield estimation. To address the issues of overlapping and small volumes of wheat heads on complex backgrounds, this paper proposes the YOLOv7-MA model. By introducing micro-scale detection layers and the convolutional block attention module, the model enhances the target information of wheat heads and weakens the background information, thereby strengthening its ability to detect small wheat heads and improving the detection performance. Experimental results indicate that after being trained and tested on the Global Wheat Head Dataset 2021, the YOLOv7-MA model achieves a mean average precision (MAP) of 93.86% with a detection speed of 35.93 frames per second (FPS), outperforming Faster-RCNN, YOLOv5, YOLOX, and YOLOv7 models. Meanwhile, when tested under the three conditions of low illumination, blur, and occlusion, the coefficient of determination (R2) of YOLOv7-MA is respectively 0.9895, 0.9872, and 0.9882, and the correlation between the predicted wheat head number and the manual counting result is stronger than others. In addition, when the YOLOv7-MA model is transferred to field-collected wheat head datasets, it maintains high performance with MAP in maturity and filling stages of 93.33% and 93.03%, respectively, and R2 values of 0.9632 and 0.9155, respectively, demonstrating better performance in the maturity stage. Overall, YOLOv7-MA has achieved accurate identification and counting of wheat heads in complex field backgrounds. In the future, its application with unmanned aerial vehicles (UAVs) can provide technical support for large-scale wheat yield estimation in the field.

Wheat Seed Detection and Counting Method Based on Improved YOLOv8 Model

“…The integration and innovation of traditional agricultural production methods with deep learning has become a general trend, and agricultural informatization and intelligentization have been vigorously developed. Currently, deep learning is widely applied in fields such as plant disease and insect pest control [ 21 , 22 , 23 ], plant counting [ 24 , 25 , 26 , 27 ], and plant phenotyping [ 28 , 29 , 30 , 31 ].…”

Lightweight YOLOv8 for Wheat Head Detection

“…However, these deep-learning detection techniques are highly dependent on features such as color and texture, and thus are susceptible to noise factors (e.g., lighting conditions, shooting angles, wheat head colors, and environmental variations), making accurate identification challenging. Owing to their limited generalizability, achieving reliable detection and counts under real field conditions has proven difficult [20]. In recent years, with the continuous development of computer vision technology and convolutional neural networks (CNNs), algorithms that monitor crop growth while accurately estimating wheat yield have been produced.…”

“…Bao et al [55] developed a method based on the transformer prediction head YOLO, optimizing its model structure and introducing coordinate attention to improve its wheat spike-counting accuracy using drone imagery. Meng et al [20] combined microscale detection layers and a CBAM in a YOLOv7 model to achieve accurate wheat spike identification and counting on complex field backgrounds. Based on this literature review, several challenges persist in wheat spike detection…”

Wheat Teacher: A One-Stage Anchor-Based Semi-Supervised Wheat Head Detector Utilizing Pseudo-Labeling and Consistency Regularization Methods

“…Zhao et al [17] improved wheat head feature extraction by adding micro-scale detection layers and setting prior anchor boxes, optimizing the YOLOv5 network structure, significantly enhancing the detection accuracy of wheat head images captured by drones. Meng et al [18] constructed an improved YOLOv7 model by incorporating attention modules, greatly enhancing wheat head detection effectiveness, and exploring the influence of attention module quantity and placement on the model. The primary research direction of the aforementioned studies can be summarized as augmenting various specially designed modules onto existing general object detectors according to the characteristics of the target objects to be detected, thereby achieving better detection performance.…”