A review of computer vision technologies for plant phenotyping
Plant phenotype plays an important role in genetics, botany, and agronomy, while the currently popular methods for phenotypic trait measurement have some limitations in aspects of cost, performance, and space-time coverage. With the rapid development of imaging technology, computing power, and algorithms, computer vision has thoroughly revolutionized the plant phenotyping and is now a major tool for phenotypic analysis. Based on the above reasons, researchers are devoted to developing image-based plant phenotyping methods as a complementary or even alternative to the manual measurement. However, the use of computer vision technology to analyze plant phenotypic traits can be affected by many factors such as research environment, imaging system, research object, feature extraction, model selection, and so on. Currently, there is no review paper to compare and analyze these methods thoroughly. Therefore, this review introduces the typical plant phenotyping methods based on computer vision in detail, with their principle, applicable range, results, and comparison. This paper extensively reviews 200+ papers of plant phenotyping in the light of its technical evolution, spanning over twenty years (from 2000 to 2020). A number of topics have been covered in this paper, including imaging technologies, plant datasets, and state-of-the-art phenotyping methods. In this review, we categorize the plant phenotyping into two main groups: plant organ phenotyping and whole-plant phenotyping. Furthermore, for each group, we analyze each research of these groups and discuss the limitations of the current approaches and future research directions.
期刊:
Computers and Electronics in Agriculture
2020
作者:
Guangyao Li,Yaru Chen,Meng Li,Ruohao Guo,Zhenbo Li
DOI:10.1016/j.compag.2020.105672
基于可见光谱的鱼苗体长估测方法研究
在鱼苗养殖过程中,同一养殖池会出现个体大的鱼苗攻击个体小的鱼苗,个体小的鱼苗会出现伤病甚至死亡,造成经济损失,鱼苗分塘和售卖价格主要与其体长参数相关,因此需要对不同大小的鱼苗进行分离。鱼苗分类主要依赖于不同大小的网筛,费时费力,且容易对鱼苗造成损伤。针对传统人工分离方法效率低下并且缺乏科学指导的问题,本文提出了基于可见光谱的鱼苗体长估测方法研究,能够根据鱼苗图像计算鱼苗长度并进行分类。为了精确无损的获取鱼苗的体长,提出了基于迁移学习ResNet50模型的鱼苗体长估测方法。首先采集在同等高度条件下拍摄的不同长度鱼苗图像,同时手工测量鱼苗的实际长度作为数据集的标签,用四种迁移学习模型AlexNet, VGG16, GoogLeNet, ResNet50对鱼苗体长进行估算,通过验证集准确率,测试集准确率,以及不同方法的运行时间三个指标进行分析, AlexNet模型验证集准确率90.04%,测试集准确率89.82%,运行时间52 min 3 s; VGG16模型验证集准确率91.01%,测试集准确率91.17%,运行时间131 min 37 s; GoogLeNet模型验证集准确率88.02%,测试集准确率88.39%,运行时间45 min 2 s; ResNet50模型验证集准确率91.92%,测试集准确率91.09%,运行时间99 min 17 s;确定方法ResNet50。该模型具有50层的Residual Network架构,用迁移学习的方法将在ImageNet上训练得到的卷积层的参数传递到训练所使用的模型上,并调整softmax层适应本文问题。对来自10种不同长度的6 677个样本的鱼苗数据集上的实验结果表明该方法可以有效地用于鱼苗分类,通过对模型ResNet50的迁移学习的层数,迭代次数,学习率,最小批处理尺寸（Mini Batch Size）进行微调以优化模型。实验结果表明,当迁移学习模型的迁移层数为30,迭代次数为6,学习率为0.001, Mini Batch Size为10时,方法效果达到最优,模型的验证集准确率94.31%,测试集的准确率达到93.93%。该算法与传统的图像处理方法相比估算鱼苗体长准确率提高2%左右。在未来实际生产场景中,可以将该方法嵌套入鱼苗体长分离装置之中,真正的做到将科研落地,投入到实际的生产之中,减少鱼苗损伤,为未来的无人渔场奠定基础。
期刊:
光谱学与光谱分析
2020
作者:
李光耀,彭芳,钮冰姗,李振波
DOI:CNKI:SUN:GUAN.0.2020-04-048
Shellfish Detection Based on Fusion Attention Mechanism in End-to-End Network
Object detection has many difficulties and challenges in the agricultural field, mainly due to the lack of data and the complexity of the agricultural environment. Therefore, we built a shellfish dataset containing 3772 images in 7 categories, all of which were manually labeled and verified. In addition, based on the SSD model framework, we used the lightweight MobileNet-v2 classification network to replace the original VGG16 network, and introduced a residual attention mechanism between the classification network and the prediction convolution layer. This could not only lead to a better capture the local features of the images, but also meet the needs of real-time and mobile use. The experimental results show that the performance of our model on the shellfish dataset is better than the current mainstream target detection models. And the verification results achieved an accuracy of 95.38% and a detection speed of 33 ms per picture, indicating that the validity of the model we proposed.
期刊:
Pattern Recognition and Computer Vision
2019
作者:
Jun Yue,Yaodong Li,Chuyue Zhang,Zhenbo Li,Guangyao Li
DOI:10.1007/978-3-030-31726-3_44
Sea Cucumber Image Dehazing Method by Fusion of Retinex and Dark Channel
This paper proposes a method based on the prior fusion of Retinex and dark channel to enhance the defogging of underwater sea cucumber images. Firstly, the original RGB image is pre-processed by dark channel prior, and then the reflection property of the image is preserved by weighted average of pixels in the image. Then, the original image is convolved with a Gaussian template to generate a high-frequency enhanced image. Finally, the brightness and saturation of the image are enhanced by changing the values of S and V in the HSV image. The experimental results are represented by four evaluation indicators such as the MSE. By processing images of sea cucumber, we obtained MSE, ENL, EI, and SNR values of 1.9782, 14.4049, 6.9586, and 14.9172, respectively. Compared with other methods, the image processed by our method has better performance in evaluating indicators. It shows that our method shows great performance in the defogging and enhancement of underwater sea cucumber images.
期刊:
IFAC-PapersOnLine
2018
作者:
Fang Peng,Bingshan Niu,Guangyao Li,Zhenbo Li
DOI:10.1016/j.ifacol.2018.08.098
Optimized BP neural network for Dissolved Oxygen prediction
To solve the low accuracy, slow convergence and poor robustness problem of traditional neural network method for water quality forecasting, a new model of dissolved oxygen content prediction is proposed based on sliding window, particle swarm optimization (PSO) and BP neural network. dissolved oxygen content prediction model in water quality is established by handling dissolved oxygen content data through sliding window, and using particle swarm optimization algorithm to obtain BP neural network parameters. This model is applied to prediction analysis of dissolved oxygen with online monitoring of regional groundwater in Xilin Gol League on July 25, 2017 to December 5, 2017. Experimental results show that the model has better prediction effect, and mean square error (MSE), root mean square error(RMSE), mean absolute error(MAE) value of PSO algorithm to optimize the BP neural network based on sliding window are 0.437% and 6.611%, 0. 251% respectively, which are better than single forecasting method by using sliding window, PSO, and BP neural network individually. The Optimized BP neural network not only has fast convergence speed and high prediction accuracy, but also provides decision-making basis for water pollution control and water management.
期刊:
IFAC-PapersOnLine
2018
作者:
Fang Peng,Bingshan Niu,Guangyao Li,Ling Zhu,Zhenbo Li,Jing Wu
DOI:10.1016/j.ifacol.2018.08.132
Survey of Fish Behavior Analysis by Computer Vision
Assessment of the behavior or physiology of cultured fish has always been difficult due to the sampling time, differences between experimental and aquaculture conditions, and methodological bias inherent. Recent developments in computer vision technology, however, have opened possibilities to better observe fish behavior. Such technology allows for non-destructive, rapid, economic, consistent, and objective inspection tools, while providing evaluation techniques based on image analysis and processing in a wide variety of applications. “Fish”, in this study, refers to underwater vertebrate fish belonging to the Pisces class that inhabit almost all available aquatic environments. This study aims to assess current, worldwide fish behavior study methods that use cameras which utilize computer vision. The evolution of computer vision as applied to fish behavior is explored in this paper for all stages of production, from hatcheries to harvest. Computer vision technology is regarded as existing from 1973 to 2018, specifically the Elsevier database. Fish behavior and underwater habitats are explored at large, especially in aquaculture fishing. Based on the methods observed above, relevant viewpoints on the present situation are presented as well as suggestions for future research directions.
期刊:
Journal of Aquaculture Research & Development
2018
作者:
Zhenbo Li,Long Zhang,Jing Wu,Fang Peng,Guangyao Li,Bingshan Niu
DOI:10.4172/2155-9546.1000534
Classification of Peanut Images Based on Multi-features and SVM
This article provides a method for accurate classification of peanuts. Peanuts can be classified into three categories, including one peanut, two peanuts and three peanuts. Because different peanuts have different prices. The characteristics of peanut images were extracted by three different methods including the convolution neural network of aspect ratio, HOG and Hu invariant moment, and then classifying peanut images respectively by the SVM (support vector machine). The accuracy rate of the aspect ratio + SVM algorithm, HOG+SVM algorithm, Hu invariant moment +SVM algorithm respectively is 96.72%, 81.97% and 81.97%, realize the industrialization of peanut classification.
期刊:
IFAC-PapersOnLine
2018
作者:
Jing Wu,Zhaolu Yang,Guangyao Li,Fang Peng,Bingshan Niu,Zhenbo Li
DOI:10.1016/j.ifacol.2018.08.110
Water Quality Prediction Model Combining Sparse Auto-encoder and LSTM Network
In order to improve the prediction accuracy of dissolved oxygen in aquaculture, a hybrid model based on sparse auto-encoder (SAE) and long-short-term memory network (LSTM) is proposed in this paper. The hidden layer data pre-trained by SAE contains deep latent features of water quality, and then input it into the LSTM to enhance the prediction accuracy. Experimental results show that SAE-LSTM surpasses LSTM through reducing MSE respectively by 23.3%, 53.6%, and 39.2% in the prediction steps of 3, 6, and 12 hours, and surpasses SAE-BPNN by 87.7%, 91.9%, and 90.0%, proving that our hybrid model is more accurate.
期刊:
IFAC-PapersOnLine
2018
作者:
Zheng Miao,Jing Wu,Guangyao Li,Bingshan Niu,Fang Peng,Zhenbo Li
DOI:10.1016/j.ifacol.2018.08.091
