There is an increasing demand for remote monitoring and management of agricultural machinery operation in conservation tillage. Considering the problems of large errors in detecting operation quality parameters, such as tillage depth and corn straw cover rate, in complex farmland environments, this paper proposes a tillage depth measurement method based on the dual attitude compound of a tractor body and three-point hitch mechanism with lower pull rod and an online measurement method based on K-means clustering of the corn straw cover rate on farmland surface. An operation monitoring terminal was developed for the remote collection of quality parameters of conservation tillage field operation. A remote monitoring system of agricultural machinery operation was constructed and applied over a large area. The field tests showed that the static mean error and root-mean-square error of this method were 0.16 and 0.67 cm for uphill and 0.36 and 0.57 cm for downhill, respectively. For the 28 and 33 cm tillage depth tests, the mean dynamic measurement errors of this method were 0.55 and 0.61 cm, and the root means square errors were 0.64 and 0.73 cm, respectively, and the coefficient of variation of tillage depth did not exceed 3%. The correlation coefficient between the corn straw cover rate detection algorithm based on K-means clustering and the manual image marking method reached 0.92, with an average error of 9.69%, and the accuracy filled the demand for straw cover rate detection. The detection accuracy of tillage depth and straw cover rate was high and thus provides an effective means of information technology support for the quality monitoring and production management of conservation tillage farming operations.

For self-driving agricultural vehicles, the sensing of the headland environment based on image recognition is an important technological aspect. Cropland headland environments are complex and diverse. Traditional image feature extraction methods have many limitations. This study proposed a method of automatic cropland headland image recognition based on deep learning. Based on the characteristics of cropland headland environments and practical application needs, a dataset was constructed containing six categories of annotated cropland headland images and an augmented headland image training set was used to train the compact network MobileNetV2. Under the same experimental conditions, the model prediction accuracy for the first ranked category in all the results (Top-1 accuracy) of the MobileNetV2 network on the validation set was 98.5%. Compared with classic ResNetV2-50, Inception-V3, and backend-compressed Inception-V3, MobileNetV2 has a high accuracy, high recognition speed, and a small memory footprint. To further test the performance of the model, 250 images were used for each of the six categories of headland images as the test set for the experiments. The average of the harmonic mean of precision and recall (F1-score) of the MobileNetV2 network for all the categories of headland images reached 97%. The MobileNetV2 network exhibits good robustness and stability. The results of this study indicate that onboard computers on self-driving agricultural vehicles are able to employ the MobileNetV2 network for headland image recognition to meet the application requirements of headland environment sensing. Keywords: cropland image, deep learning, image recognition, model compression, MobileNetV2 network DOI: 10.25165/j.ijabe.20231602.6195 Citation: Qiao Y J, Liu H, Meng Z J, Chen J P, Ma L Y. Method for the automatic recognition of cropland headland images based on deep learning. Int J Agric & Biol Eng, 2023; 16(2): 216-224.

Field-scale information on rice seedling distribution is crucial for transplanting quality assessment, site-specific seedling-stage management, and precision agricultural operations. However, seedlings in UAV orthomosaics are small, closely spaced, and easily confused with soil, shallow water, plastic mulch, and shadows, which makes accurate localization challenging. Meanwhile, under high-density planting, existing segmentation-based networks often generate adhesive responses between neighboring seedlings, leading to missed detections and biased counting. To address these issues, this study proposed PPNet, a deep learning framework for rice seedling localization and distribution mapping from UAV orthomosaics. The framework uses root-based point annotation and integrates a rice-point perception module (RPPM), an adjacent decoupling module (ADM), and an aggregated attention module (AAM). These modules enhance seedling-point responses, separate adjacent seedlings, and introduce row-structure information. Experiments were conducted on a UAV orthomosaic dataset containing 3,013 image tiles and 220,553 seedling point annotations from different rice-growing regions. PPNet achieved an F1 score of 0.9029, an AP of 0.8881, a counting MAE of 6.99, and an R2 of 0.9553. The method also generated field-scale seedling point maps and density maps under different planting densities and field backgrounds. The results demonstrate that PPNet can provide reliable seedling localization and counting information for crop establishment assessment and targeted seedling management, supporting intelligent agricultural machinery decision-making in precision farming operations.