Performance of deep learning models for classifying and detecting common weeds in corn and soybean production systems

Ahmad, Aanis; Saraswat, Dharmendra; Aggarwal, Varun; Etienne, Aaron; Hancock, Benjamin

doi:10.1016/j.compag.2021.106081

Cited by 93 publications

(49 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…For example, Xiong et al [9] adopted the YOLO v2 (you only look once v2) network to estimate the number of mangos. Moreover, YOLO v3 had been widely used for the recognition of lychee fruits and stems [10], weeds [11], tea shoots [12], coffee fruits [13], and cows [14][15][16]. The above findings showed that the YOLO [17][18][19] series network had achieved good results when processing images in the biosystems engineering domain, which reduces manual pre-processing and post-processing steps.…”

Section: Introductionmentioning

confidence: 88%

Using Channel and Network Layer Pruning Based on Deep Learning for Real-Time Detection of Ginger Images

Fang

et al. 2021

Agriculture

View full text Add to dashboard Cite

Consistent ginger shoot orientation helps to ensure consistent ginger emergence and meet shading requirements. YOLO v3 is used to recognize ginger images in response to the current ginger seeder’s difficulty in meeting the above agronomic problems. However, it is not suitable for direct application on edge computing devices due to its high computational cost. To make the network more compact and to address the problems of low detection accuracy and long inference time, this study proposes an improved YOLO v3 model, in which some redundant channels and network layers are pruned to achieve real-time determination of ginger shoots and seeds. The test results showed that the pruned model reduced its model size by 87.2% and improved the detection speed by 85%. Meanwhile, its mean average precision (mAP) reached 98.0% for ginger shoots and seeds, only 0.1% lower than the model before pruning. Moreover, after deploying the model to the Jetson Nano, the test results showed that its mAP was 97.94%, the recognition accuracy could reach 96.7%, and detection speed could reach 20 frames·s−1. The results showed that the proposed method was feasible for real-time and accurate detection of ginger images, providing a solid foundation for automatic and accurate ginger seeding.

show abstract

Section: Introductionmentioning

confidence: 88%

Using Channel and Network Layer Pruning Based on Deep Learning for Real-Time Detection of Ginger Images

Fang

et al. 2021

Agriculture

View full text Add to dashboard Cite

show abstract

“…TN (True Negative) represents the number of samples whose true value was negative and predicted value was positive, where FN and FP are the first and second types of error. Theoretically, the larger the TP and TN, the more accurate the model, and the smaller the FP and FN, the better the performance [25,26]. and 30% for verification; they were randomly rotated −90° to 90° and also randomly zoomed in and out by 1 to 2 times.…”

Section: Machine Learning and Model Performance Indexmentioning

confidence: 99%

“…The closer to 1, the better the model, and the closer to 0, the worse the model. The extended classification model performance indicators for the confusion matrix are as follows [25]:…”

Section: 𝐴𝑐𝑐 = 𝑇𝑃 + 𝑇𝑁 𝑇𝑃 + 𝐹𝑃 + 𝐹𝑁 + 𝑇𝑁mentioning

confidence: 99%

Identification and Machine Learning Prediction of Nonlinear Behavior in a Robotic Arm System

Wang

Zhu

2021

Symmetry

View full text Add to dashboard Cite

In this study, the subject of investigation was the dynamic double pendulum crank mechanism used in a robotic arm. The arm is driven by a DC motor though the crank system and connected to a fixed side with a mount that includes a single spring and damping. Robotic arms are now widely used in industry, and the requirements for accuracy are stringent. There are many factors that can cause the induction of nonlinear or asymmetric behavior and even excite chaotic motion. In this study, bifurcation diagrams were used to analyze the dynamic response, including stable symmetric orbits and periodic and chaotic motions of the system under different damping and stiffness parameters. Behavior under different parameters was analyzed and verified by phase portraits, the maximum Lyapunov exponent, and Poincaré mapping. Firstly, to distinguish instability in the system, phase portraits and Poincaré maps were used for the identification of individual images, and the maximum Lyapunov exponents were used for prediction. GoogLeNet and ResNet-50 were used for image identification, and the results were compared using a convolutional neural network (CNN). This widens the convolutional layer and expands pooling to reduce network training time and thickening of the image; this deepens the network and strengthens performance. Secondly, the maximum Lyapunov exponent was used as the key index for the indication of chaos. Gaussian process regression (GPR) and the back propagation neural network (BPNN) were used with different amounts of data to quickly predict the maximum Lyapunov exponent under different parameters. The main finding of this study was that chaotic behavior occurs in the robotic arm system and can be more efficiently identified by ResNet-50 than by GoogLeNet; this was especially true for Poincaré map diagnosis. The results of GPR and BPNN model training on the three types of data show that GPR had a smaller error value, and the GPR-21 × 21 model was similar to the BPNN-51 × 51 model in terms of error and determination coefficient, showing that GPR prediction was better than that of BPNN. The results of this study allow the formation of a highly accurate prediction and identification model system for nonlinear and chaotic motion in robotic arms.

show abstract

“…Security applications such as facial recognition [12], pedestrian avoidance [13], and obstacle avoidance [14] also rely on computer vision. Although computer vision is commonly used, its recent implementation in precision agriculture applications has shown promising results for the detection of different stresses within crop fields, such as weeds [15], diseases [16], pests [17], nutrient deficiencies [18], etc. In addition, it has been used for fruit counting [19], crop height detection [20], automation [21], and assessment of fruit and vegetable quality [22].…”

Section: Introductionmentioning

confidence: 99%

“…Therefore, training deep learning models gained popularity as they rely on convolutional neural networks capable of automatically extracting important features from images [31]. Deep learning was recently used for weed identification in corn using the You Only Look Once (YOLOv3) algorithm [15].…”

Section: Introductionmentioning

confidence: 99%

Deep Learning-Based Object Detection System for Identifying Weeds Using UAS Imagery

et al. 2021

Self Cite

View full text Add to dashboard Cite

Current methods of broadcast herbicide application cause a negative environmental and economic impact. Computer vision methods, specifically those related to object detection, have been reported to aid in site-specific weed management procedures for targeted herbicide application within a field. However, a major challenge to developing a weed detection system is the requirement for a properly annotated database to differentiate between weeds and crops under field conditions. This research involved creating an annotated database of 374 red, green, and blue (RGB) color images organized into monocot and dicot weed classes. The images were acquired from corn and soybean research plots located in north-central Indiana using an unmanned aerial system (UAS) flown at 30 and 10 m heights above ground level (AGL). A total of 25,560 individual weed instances were manually annotated. The annotated database consisted of four different subsets (Training Image Sets 1–4) to train the You Only Look Once version 3 (YOLOv3) deep learning model for five separate experiments. The best results were observed with Training Image Set 4, consisting of images acquired at 10 m AGL. For monocot and dicot weeds, respectively, an average precision (AP) score of 91.48 % and 86.13% was observed at a 25% IoU threshold (AP @ T = 0.25), as well as 63.37% and 45.13% at a 50% IoU threshold (AP @ T = 0.5). This research has demonstrated a need to develop large, annotated weed databases to evaluate deep learning models for weed identification under field conditions. It also affirms the findings of other limited research studies utilizing object detection for weed identification under field conditions.

show abstract

Performance of deep learning models for classifying and detecting common weeds in corn and soybean production systems

Cited by 93 publications

References 27 publications

Using Channel and Network Layer Pruning Based on Deep Learning for Real-Time Detection of Ginger Images

Using Channel and Network Layer Pruning Based on Deep Learning for Real-Time Detection of Ginger Images

Identification and Machine Learning Prediction of Nonlinear Behavior in a Robotic Arm System

Deep Learning-Based Object Detection System for Identifying Weeds Using UAS Imagery

Contact Info

Product

Resources

About