Introduction: The precise detection of weeds in the field is the premise of implementing weed management. However, the similar color, morphology, and occlusion between wheat and weeds pose a challenge to the detection of weeds. In this study, a CSCW-YOLOv7 based on an improved YOLOv7 architecture was proposed to identify five types of weeds in complex wheat fields.Methods: First, a dataset was constructed for five weeds that are commonly found, namely, Descurainia sophia, thistle, golden saxifrage, shepherd’s purse herb, and Artemisia argyi. Second, a wheat weed detection model called CSCW-YOLOv7 was proposed to achieve the accurate identification and classification of wheat weeds. In the CSCW-YOLOv7, the CARAFE operator was introduced as an up-sampling algorithm to improve the recognition of small targets. Then, the Squeeze-and-Excitation (SE) network was added to the Extended Latent Attention Networks (ELAN) module in the backbone network and the concatenation layer in the feature fusion module to enhance important weed features and suppress irrelevant features. In addition, the contextual transformer (CoT) module, a transformer-based architectural design, was used to capture global information and enhance self-attention by mining contextual information between neighboring keys. Finally, the Wise Intersection over Union (WIoU) loss function introducing a dynamic nonmonotonic focusing mechanism was employed to better predict the bounding boxes of the occluded weed.Results and discussion: The ablation experiment results showed that the CSCW-YOLOv7 achieved the best performance among the other models. The accuracy, recall, and mean average precision (mAP) values of the CSCW-YOLOv7 were 97.7%, 98%, and 94.4%, respectively. Compared with the baseline YOLOv7, the improved CSCW-YOLOv7 obtained precision, recall, and mAP increases of 1.8%, 1%, and 2.1%, respectively. Meanwhile, the parameters were compressed by 10.7% with a 3.8-MB reduction, resulting in a 10% decrease in floating-point operations per second (FLOPs). The Gradient-weighted Class Activation Mapping (Grad-CAM) visualization method suggested that the CSCW-YOLOv7 can learn a more representative set of features that can help better locate the weeds of different scales in complex field environments. In addition, the performance of the CSCW-YOLOv7 was compared to the widely used deep learning models, and results indicated that the CSCW-YOLOv7 exhibits a better ability to distinguish the overlapped weeds and small-scale weeds. The overall results suggest that the CSCW-YOLOv7 is a promising tool for the detection of weeds and has great potential for field applications.

The rice spike, a crucial part of rice plants, plays a vital role in yield estimation, pest detection, and growth stage management in rice cultivation. When using drones to capture photos of rice fields, the high shooting angle and wide coverage area can cause rice spikes to appear small in the captured images and can cause angular distortion of objects at the edges of images, resulting in significant occlusions and dense arrangements of rice spikes. These factors are unique challenges during drone image acquisition that may affect the accuracy of rice spike detection. This study proposes a rice spike detection method that combines deep learning algorithms with drone perspectives. Initially, based on an enhanced version of YOLOv5, the EMA (efficient multiscale attention) attention mechanism is introduced, a novel neck network structure is designed, and SIoU (SCYLLA intersection over union) is integrated. Experimental results demonstrate that RICE-YOLO achieves a mAP@0.5 of 94.8% and a recall of 87.6% on the rice spike dataset. During different growth stages, it attains an AP@0.5 of 96.1% and a recall rate of 93.1% during the heading stage, and a AP@0.5 of 86.2% with a recall rate of 82.6% during the filling stage. Overall, the results indicate that the proposed method enables real-time, efficient, and accurate detection and counting of rice spikes in field environments, offering a theoretical foundation and technical support for real-time and efficient spike detection in the management of rice growth processes.

Leaf chlorophyll content is crucial for monitoring plant growth and photosynthetic capacity. The Soil and Plant Analysis Development (SPAD) values are widely utilized as a relative chlorophyll content index in ecological agricultural surveys and vegetation remote sensing applications. Multi-spectral cameras are a cost-effective alternative to hyperspectral cameras for agricultural monitoring. However, the limited spectral bands of multi-spectral cameras restrict the number of vegetation indices (VIs) that can be synthesized, necessitating the exploration of other options for SPAD estimation. This study evaluated the impact of using texture indices (TIs) and VIs, alone or in combination, for estimating rice SPAD values during different growth stages. A multi-spectral camera was attached to an unmanned aerial vehicle (UAV) to collect remote sensing images of the rice canopy, with manual SPAD measurements taken immediately after each flight. Random forest (RF) was employed as the regression method, and evaluation metrics included coefficient of determination ($R^2$) and root mean squared error ($RMSE$). The study found that textural information extracted from multi-spectral images could effectively assess the SPAD values of rice. Constructing TIs by combining two textural feature values (TFVs) further improved the correlation of textural information with SPAD. Utilizing both VIs and TIs demonstrated superior performance throughout all growth stages. The model works well in estimating the rice SPAD in an independent experiment in 2022, proving that the model has good generalization ability. The results suggest that incorporating both spectral and textural data can enhance the precision of rice SPAD estimation throughout all growth stages, compared to using spectral data alone. These findings are of significant importance in the fields of precision agriculture and environmental protection.

Monitoring the performance of hybrid rice seeding is very important for the seedling production line to adjust the sowing amount of the seeding device. The objective of this paper was to develop a system for the real-time online monitoring of the performance of hybrid rice seeding based on embedded machine vision and machine learning technology. The embedded detection system captured images of pot trays that passed under the illuminant cabinet installed in the seedling production line. This paper proposed an algorithm for fixed threshold segmentation by analyzing the images with the exploratory analysis method. With the algorithm, the grid image and seed image were extracted from the pot tray image. The paper also proposed a method for obtaining pixel coordinates of gridlines from the grid image. Binary images of seeds were divided into small pieces, according to the pixel coordinates of gridlines. Each piece corresponded to a cell on the pot tray. By scanning the contours in each piece of the image to check whether there were seeds in the cell, the number of empty cells was counted and then used to calculate the missing rate of hybrid rice seeding. The seed number sowed in pot trays was monitored while using the machine learning approach. The experimental results demonstrated that it would consume 4.863 s for the device to process an image, which allowed for the detection of the missing rate and seed number in real-time at the rate of 500 trays per hour (7.2 s per tray). The average accuracy of the detection of missing rates of a seedling production line was 94.67%. The average accuracy of the detection of seed numbers was 95.68%.

Accurately detecting rice panicles in complex field environments remains challenging due to their small size, dense distribution, diverse growth directions, and easy confusion with the background. To accurately detect rice panicles, this study proposes OE-YOLO, an enhanced framework derived from YOLOv11, incorporating three synergistic innovations. First, oriented bounding boxes (OBB) replace horizontal bounding boxes (HBB) to precisely capture features of rice panicles across different heights and growth stages. Second, the backbone network is redesigned with EfficientNetV2, leveraging its compound scaling strategy to balance multi-scale feature extraction and computational efficiency. Third, a C3k2_DConv module improved by dynamic convolution is introduced, enabling input-adaptive kernel fusion to amplify discriminative features while suppressing background interference. Extensive experiments on rice Unmanned Aerial Vehicle (UAV) imagery demonstrate OE-YOLO’s superiority, achieving 86.9% mAP50 and surpassing YOLOv8-obb and YOLOv11 by 2.8% and 8.3%, respectively, with only 2.45 M parameters and 4.8 GFLOPs. The model has also been validated at flight heights of 3 m and 10 m and during the heading and filling stages, achieving mAP50 improvements of 8.3%, 6.9%, 6.7%, and 16.6% compared to YOLOv11, respectively, demonstrating the generalization capability of the model. These advancements demonstrated OE-YOLO as a computationally frugal yet highly accurate solution for real-time crop monitoring, addressing critical needs in precision agriculture for robust, oriented detection under resource constraints.

To optimize rice yield and enhance quality through targeted field management at each growth stage, rapid and accurate identification of rice growth stages is crucial. This study presents the Mobilenetv3-YOLOv8 rice growth-stage recognition model, designed for high efficiency and accuracy using Unmanned Aerial Vehicle (UAV) imagery. A UAV captured images of rice fields across five distinct growth stages from two altitudes (3 m and 20 m) across two independent field experiments. These images were processed to create training, validation, and test datasets for model development. Mobilenetv3 was introduced to replace the standard YOLOv8 backbone, providing robust small-scale feature extraction through multi-scale feature fusion. Additionally, the Coordinate Attention (CA) mechanism was integrated into YOLOv8’s backbone, outperforming the Convolutional Block Attention Module (CBAM) by enhancing position-sensitive information capture and focusing on crucial pixel areas. Compared to the original YOLOv8, the enhanced Mobilenetv3-YOLOv8 model improved rice growth-stage identification accuracy and reduced the computational load. With an input image size of 400 × 400 pixels and the CA implemented in the second and third backbone layers, the model achieved its best performance, reaching 84.00% mAP and 84.08% recall. The optimized model achieved parameters and Giga Floating Point Operations (GFLOPs) of 6.60M and 0.9, respectively, with precision values for tillering, jointing, booting, heading, and filling stages of 94.88%, 93.36%, 67.85%, 78.31%, and 85.46%, respectively. The experimental results revealed that the optimal Mobilenetv3-YOLOv8 shows excellent performance and has potential for deployment in edge computing devices and practical applications for in-field rice growth-stage recognition in the future.