Efficient detection and evaluation of soybean seedling emergence is an important measure for making field management decisions. However, there are many indicators related to emergence, and using multiple models to detect them separately makes data processing too slow to aid timely field management. In this study, we aimed to integrate several deep learning and image processing methods to build a model to evaluate multiple soybean seedling emergence information. An unmanned aerial vehicle (UAV) was used to acquire soybean seedling RGB images at emergence (VE), cotyledon (VC), and first node (V1) stages. The number of soybean seedlings that emerged was obtained by the seedling emergence detection module, and image datasets were constructed using the seedling automatic cutting module. The improved AlexNet was used as the backbone network of the growth stage discrimination module. The above modules were combined to calculate the emergence proportion in each stage and determine soybean seedlings emergence uniformity. The results show that the seedling emergence detection module was able to identify the number of soybean seedlings with an average accuracy of 99.92%, a R² of 0.9784, a RMSE of 6.07, and a MAE of 5.60. The improved AlexNet was more lightweight, training time was reduced, the average accuracy was 99.07%, and the average loss was 0.0355. The model was validated in the field, and the error between predicted and real emergence proportions was up to 0.0775 and down to 0.0060. It provides an effective ensemble learning model for the detection and evaluation of soybean seedling emergence, which can provide a theoretical basis for making decisions on soybean field management and precision operations and has the potential to evaluate other crops emergence information.

The extraction of navigation lines is a crucial aspect in the field autopilot system for intelligent agricultural equipment. Given that soybean seedlings are small, and straw can be found in certain Northeast China soybean fields, accurately obtaining feature points and extracting navigation lines during the soybean seedling stage poses numerous challenges. To solve the above problems, this paper proposes a method of extracting navigation lines based on the average coordinate feature points of pixel points in the bean seedling belt according to the calculation of the average coordinate. In this study, the soybean seedling was chosen as the research subject, and the Hue, Saturation, Value (HSV) colour model was employed in conjunction with the maximum interclass variance (OTSU) method for RGB image segmentation. To extract soybean seedling bands, a novel approach of framing binarised image contours by drawing external rectangles and calculating average coordinates of white pixel points as feature points was proposed. The feature points were normalised, and then the improved adaptive DBSCAN clustering method was used to cluster the feature points. The least squares method was used to fit the centre line of the crops and the navigation line, and the results showed that the average distance deviation and the average angle deviation of the proposed algorithm were 7.38 and 0.32. The fitted navigation line achieved an accuracy of 96.77%, meeting the requirements for extracting navigation lines in intelligent agricultural machinery equipment for soybean inter-row cultivation. This provides a theoretical foundation for realising automatic driving of intelligent agricultural machinery in the field.

Accurate navigation line recognition in mature soybean fields presents significant challenges due to complex backgrounds. To address this issue, we developed an enhanced YOLOv10n-based model for robust visual navigation, and the assessment was conducted in the experimental fields of the laboratory. The dataset comprised 1363 original images collected at this site and was expanded to 5452 images through data augmentation. The study utilized an innovative data annotation approach focusing on inter-ridge navigation areas to minimize background noise from mature soybean rows. The model was optimized by integrating the CSP Multi-Scale Edge Information Enhancement (CSP-MEIE) module and a lightweight detection head. This architecture significantly improves efficiency, achieving a model size of just 4.5 MB and a parameter count of 2.137 M, while delivering a rapid detection speed of 204.1 FPS. Crucially, the model expands the effective receptive field to 96.6% (t = 99%), far exceeding the 73.0% of the baseline YOLOv10n, ensuring robust feature capture without compromising accuracy (92.6% mAP50-95). For path planning, path points were extracted and predicted using a combination of Kalman filtering and adaptive segmentation. Field experiments demonstrated the system’s effectiveness, achieving an average distance error of 4.53 pixels and an average angle error of 3.57°, a processing speed of 28.17 ms per frame, and a navigation line recognition accuracy of 98.05%. These findings highlight the method’s capability to meet real-time agricultural requirements, offering a reliable visual perception and decision-making basis for autonomous navigation in complex field environments.