Multi-label learning for crop leaf diseases recognition and severity estimation based on convolutional neural networks
- PDF / 3,026,609 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 52 Downloads / 215 Views
METHODOLOGIES AND APPLICATION
Multi-label learning for crop leaf diseases recognition and severity estimation based on convolutional neural networks Miaomiao Ji2 · Keke Zhang2 · Qiufeng Wu1 · Zhao Deng2
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Crop diseases have always been a dilemma as it can cause significant diminution in both quality and quantity of agricultural yields. Thus, automatic recognition and severity estimation of crop diseases on leaves plays a crucial role in agricultural sector. In this paper, we propose a series of automatic image-based crop leaf diseases recognition and severity estimation networks, i.e., BR-CNNs, which can simultaneously recognize crop species, classify crop diseases and estimate crop diseases severity based on deep learning. BR-CNNs based on binary relevance (BR) multi-label learning algorithm and deep convolutional neural network (CNN) approaches succeed in identifying 7 crop species, 10 crop diseases types (including Healthy) and 3 crop diseases severity kinds (normal, general and serious). Compared with LP-CNNs and MLP-CNNs, the overall performance of BR-CNNs is superior. The BR-CNN based on ResNet50 achieves the best test accuracy of 86.70%, which demonstrates the feasibility and effectiveness of our network. The BR-CNN based on the light-weight NasNet also achieves excellent test accuracy of 85.28%, which can provide more possibilities for the development of mobile systems and devices. Keywords Multi-label · Crop diseases recognition · Crop diseases severity estimation · Convolutional neural network · Computer vision
1 Introduction Crop diseases have grown to be a dilemma as they can cause significant reduction in both quality and quantity of agricultural yields. It was estimated that 2015 crop diseases loss in Georgia (USA) was approximately $740.86 million. Of this amount, around $233.57 million was spent on controlling the diseases which accounted for 31.52%, and the rest was damage value caused by crop diseases (Hiary et al. 2011). Those statistics are listed in Table 1. Lagging diagnosis and incorrect treatment bear responsibility for a significant portion of damage loss and controlling costs. A very small number of diseased leaves can spread the infection to the Communicated by V. Loia.
B
Qiufeng Wu [email protected] Miaomiao Ji [email protected]
1
College of Science, Northeast Agricultural University, Harbin 150030, China
2
College of Engineering, Northeast Agricultural University, Harbin 150030, China
whole batch of crops and cause epidemic all over the field, which is undoubtedly devastating. There is therefore a need for agricultural producers to obtain the crop diseases hazard information as soon as possible and adopt more accurate treatment measures, such as precise targeted therapy and reactive pesticide dosage strategies depending on the harm degree of crop diseases to avoid excessive economical cost. Apart from monitoring epidemics, accurate assessment of diseases severity is also critical for evaluation of germ pl
Data Loading...
