Improved Real-Time Semantic Segmentation Network Model for Crop Vision Navigation Line Detection

Abstract

Field crops are generally planted in rows to improve planting efficiency and facilitate field management. Therefore, automatic detection of crop planting rows is of great significance for achieving autonomous navigation and precise spraying in intelligent agricultural machinery and is an important part of smart agricultural management. To study the visual navigation line extraction technology of unmanned aerial vehicles (UAVs) in farmland environments and realize real-time precise farmland UAV operations, we propose an improved ENet semantic segmentation network model to perform row segmentation of farmland images. Considering the lightweight and low complexity requirements of the network for crop row detection, the traditional network is compressed and replaced by convolution. Based on the residual network, we designed a network structure of the shunting process, in which low-dimensional boundary information in the feature extraction process is passed backward using the residual stream, allowing efficient extraction of low-dimensional information and significantly improving the accuracy of boundary locations and row-to-row segmentation of farmland crops. According to the characteristics of the segmented image, an improved random sampling consensus algorithm is proposed to extract the navigation line, define a new model-scoring index, find the best point set, and use the least-squares method to fit the navigation line. The experimental results showed that the proposed algorithm allows accurate and efficient extraction of farmland navigation lines, and it has the technical advantages of strong robustness and high applicability. The algorithm can provide technical support for the subsequent quasi-flight of agricultural UAVs in farmland operations.

Row Detection BASED Navigation and Guidance for Agricultural Robots and Autonomous Vehicles in Row-Crop Fields: Methods and Applications

“…Considering the field variability of dryland crops, Ronchetti et al [177] combined the threshold segmentation algorithm, classification algorithm, and Bayesian segmentation algorithm to effectively separate crop rows from soil background and weeds, optimizing the operational management of irrigation robots and improving the quality of crop yields. To solve the problem of slow visual navigation line extraction for irrigation robots, Cao et al [163] enhanced the ENet semantic segmentation network model for the row segmentation of crop images in drylands. By designing the network structure of shunt processing and compressing the traditional ENet network, the accuracy of the beet field boundary's location and row-to-row segmentation was improved.…”

Multi-Crop Navigation Line Extraction Based on Improved YOLO-v8 and Threshold-DBSCAN under Complex Agricultural Environments

“…However, crop row detection methods based on semantic segmentation are not obvious to crop boundary information, and most of the studies converted the crop rows into a rough long rectangular bar for mask prediction, ignoring the edge information of each crop [19]. On the other hand, semantic segmentation-based crop row detection models have higher computational complexity and require larger training datasets due to the need for pixel-level annotation [20]. In addition to semantic segmentation-based approaches, the application of object detection techniques emerges as a viable alternative for effectively detecting crop rows.…”

SMR-RS: An Improved Mask R-CNN Specialized for Rolled Rice Stubble Row Segmentation

“…As a research hotspot in the field of deep learning, convolutional neural networks have achieved excellent research results in agriculture in recent years, and they are widely used in various agricultural vision applications [30][31][32][33][34] and have demonstrated higher accuracy and wider applicability than other normal algorithms [35]. Deep learning-based crop row segmentation methods have also achieved excellent results: Silva [36] et al and Cao [37] et al used the Unet [38] model, and the improved Enet [39] model implemented segmentation of crop rows from an open dataset containing images of sugar beet rows in various complex environments. Not only have they achieved accurate segmentation of the crop rows, but the former has done so with robustness to shading and different growth stages, and the latter improved boundary localization accuracy.…”