Precise sowing is one of the most fundamental procedures in modern agriculture. The quality of seed placement within the furrow can directly influence the maize yield, uniformity, and resource efficiency. Conventional seed metering monitoring systems can often rely on the optical or piezoelectric sensors that are mounted on the seed tube. Since they can only infer the seed flow from the metering device, they cannot assess the actual final quality of seed placement within the seed furrow after deposition. Some influencing factors (such as the seed bounce, furrow opener-induced soil disturbance, and covering dynamics) can also alter the seed spatial distribution, leading to the undetected miss-seeding or multi-seeding events. The compromise can also reduce the harvest potential. In this study, an integrated system was developed and then validated for the direct detection and assessment of the seeding quality in the furrow. A flexible piezoresistive sensor array was seamlessly integrated into the seed pressing tongue. There was direct contact with the seeds and soil in the closed furrow, thus ensuring intimate interaction with the seeding environment. Three primary components consisted of: the flexible piezoresistive sensor array embedded at the bottom of the seed pressing tongue, where the subtle pressure was captured when seeds passed through; a central controller for data acquisition from the real-time signals with minimal latency; and a signal processing module to interpret the raw data using advanced algorithms. The signals of the dynamic pressure were generated as the seeds went beneath the press wheel. Each seed shared a distinct pressure signature using its size, shape, and mechanical properties. The sensor array was used to record the number, duration, and intervals of the high-level output signals. The signal analysis was carried out to distinguish between the transient pressure spikes of individual seeds and the overlapping seeds, and the prolonged signals of multiple seeds in close proximity. Thereby, the misses and multiple seeding events were evaluated even in the challenging fields. An experimental protocol was executed to validate the system. Initial performance tests were conducted on the sensor to define its response curve, spatial sensitivity, and its ability to differentiate between the mechanical impedance of seeds and soil. Subsequently, a seed displacement test was carried out, where the seeds were tracked using high-speed cameras. The seeds were displaced significantly beyond their intended positions after the press wheel's action, in order to evaluate the seeding quality. Single-factor experiments were then performed to investigate the effects of the soil moisture content and forward speed on the detection accuracy, with each parameter varied while the rest was constant to isolate their impacts. Finally, a field validation test was conducted over three soil types under varying weather. The system's performance was then verified under real-world conditions, including temperature fluctuations and minor terrain undulations. The seed displacement test results show that the seed movement caused by the press wheel had a negligible impact on the quality assessment, with the average displacement at less than 5mm, fully meeting the acceptable tolerances during precise seeding. System performance shared a slight correlation with the forward speed. The accuracy decreased minimally as speed increased, likely due to the reduced signal resolution at higher velocities. Furthermore, the recognition accuracy for both single and overlapping seeds exceeded 98.0% at the lower speeds (4-6 km/h), indicating the exceptional precision. Meanwhile, the accuracy remained remarkably high (above 96.2%) even at higher operational speeds (6-10 km/h), indicating the effective function in the typical field operations. The optimal performance was achieved within a soil moisture content range of 15% to 25%, with the accuracy dropping slightly outside due to the soil's mechanical properties under pressure transmission. The system's qualification rate detection error does not exceed 0.39 percentage points at low speed and 1.9 percentage points at high speed, respectively, indicating the excellent robustness and adaptability in the varying environmental conditions. The better performance was achieved, compared with the previous indirect systems, on the assumption of seed flow. In conclusion, the system can be expected to effectively in-situ monitor the seeding quality within the seed furrow. The finding can provide a highly accurate and robust technical solution to reduce the input costs for high crop productivity, with potential applications in seed placement. A valuable reference can also offer guidance for the next-generation seed monitoring technologies in precision agriculture.

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 real-world 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.