WheatLFANet: in-field detection and counting of wheat heads with high-real-time global regression network

Abstract

Background Detection and counting of wheat heads are of crucial importance in the field of plant science, as they can be used for crop field management, yield prediction, and phenotype analysis. With the widespread application of computer vision technology in plant science, monitoring of automated high-throughput plant phenotyping platforms has become possible. Currently, many innovative methods and new technologies have been proposed that have made significant progress in the accuracy and robustness of wheat head recognition. Nevertheless, these methods are often built on high-performance computing devices and lack practicality. In resource-limited situations, these methods may not be effectively applied and deployed, thereby failing to meet the needs of practical applications. Results In our recent research on maize tassels, we proposed TasselLFANet, the most advanced neural network for detecting and counting maize tassels. Building on this work, we have now developed a high-real-time lightweight neural network called WheatLFANet for wheat head detection. WheatLFANet features a more compact encoder-decoder structure and an effective multi-dimensional information mapping fusion strategy, allowing it to run efficiently on low-end devices while maintaining high accuracy and practicality. According to the evaluation report on the global wheat head detection dataset, WheatLFANet outperforms other state-of-the-art methods with an average precision AP of 0.900 and an R2 value of 0.949 between predicted values and ground truth values. Moreover, it runs significantly faster than all other methods by an order of magnitude (TasselLFANet: FPS: 61). Conclusions Extensive experiments have shown that WheatLFANet exhibits better generalization ability than other state-of-the-art methods, and achieved a speed increase of an order of magnitude while maintaining accuracy. The success of this study demonstrates the feasibility of achieving real-time, lightweight detection of wheat heads on low-end devices, and also indicates the usefulness of simple yet powerful neural network designs.

Accurate and fast implementation of soybean pod counting and localization from high-resolution image

“…We employ CSPDarknet ( Bochkovskiy et al., 2020 ) as the backbone feature extraction part of the encoder. CSPDarknet is a widely validated and used backbone network known for its high efficiency and outstanding accuracy ( Lu et al., 2023 ; Ye et al., 2023 ; Ye and Yu, 2024 ; Ye et al., 2024 ). The entire encoder consists of 5 convolution layers and 4 feature extraction layers, specifically defined as C3(16)-C3(32)-M(32)-C3(64)-M(64)-C3(128)-M(128)-C3(256)-M(256)-S5(256).…”

“…This indicates that PodNet can run in environments with limited hardware resources and has relatively low demands on computational resources. The degree of model lightweighting is particularly important for deploying the model on lightweight platforms, as they are often constrained by computing resources and memory limitations ( Ye et al., 2023 ), such as embedded systems and Jetson Nano development boards. Especially in the field of agriculture, lightweight models can be crucial for certain agricultural managers, effectively reducing their economic burden.…”

Oriented feature pyramid network for small and dense wheat heads detection and counting

“…Based on the GWHD dataset, the WheatLFANet model proposed by Ye, J 18 is able to operate efficiently on low-end devices while maintaining high accuracy and utility. Jun S 19 proposed a WHCnet model utilizing the Augmented Feature Pyramid Networks (AugFPN) to aggregate feature information and using cascaded Intersection over Union (IoU) threshold to remove negative samples to improve the training effect, and finally using a novel detection pipeline object counting method to count wheat sheaves from the top view in the field.…”

WH-DETR: An Efficient Network Architecture for Wheat Spike Detection in Complex Backgrounds

“…Sun et al [16] proposed an improved wheat head counting network, WHCnet, employing an enhanced feature pyramid network (AugFPN) to address issues with poor wheat head detection. Ye et al [17] introduced a real-time lightweight neural network named WheatLFANet for the efficient detection and counting of wheat heads, suitable for deployment on low-end devices. Yan et al [18] developed a method for refining the scale of detection layers in a wheat spike detection network using the deep learning interpretation method GradCAM.…”

“…However, this also results in limited context capture, a drawback particularly evident in the complex scenarios of wheat spike detection. Wheat spike object detection still faces challenges such as overlap and crossing, occlusions and shadows, light transformations, changes in angle and scale, varietal differences, and growth environments [17], which hinder the performance improvement of wheat spike detection. The accuracy issues of the network, computational efficiency, and adaptability under different environmental conditions remain pressing concerns to be addressed.…”