Zhenfa Li

Bio: Zhenfa Li is an academic researcher. The author has contributed to research in topics: Canopy & Hyperspectral imaging. The author has an hindex of 2, co-authored 2 publications receiving 21 citations.

Leaf Counting with Multi-Scale Convolutional Neural Network Features and Fisher Vector Coding

Abstract: The number of leaves in maize plant is one of the key traits describing its growth conditions. It is directly related to plant development and leaf counts also give insight into changing plant development stages. Compared with the traditional solutions which need excessive human interventions, the methods of computer vision and machine learning are more efficient. However, leaf counting with computer vision remains a challenging problem. More and more researchers are trying to improve accuracy. To this end, an automated, deep learning based approach for counting leaves in maize plants is developed in this paper. A Convolution Neural Network(CNN) is used to extract leaf features. The CNN model in this paper is inspired by Google Inception Net V3, which using multi-scale convolution kernels in one convolution layer. To compress feature maps generated from some middle layers in CNN, the Fisher Vector (FV) is used to reduce redundant information. Finally, these encoded feature maps are used to regress the leaf numbers by using Random Forests. To boost the related research, a relatively single maize image dataset (Different growth stage with 2845 samples, which 80% for train and 20% for test) is constructed by our team. The proposed algorithm in single maize data set achieves Mean Square Error (MSE) of 0.32.

Nitrogen Balance Index Prediction of Winter Wheat by Canopy Hyperspectral Transformation and Machine Learning

Abstract: Nitrogen balance index (NBI) is an important indicator for scientific diagnostic and quantitative research on crop growth status. The quick and accurate assessment of NBI is necessary for farmers to make timely N management decisions. The objective of the study was to estimate winter wheat NBI based on canopy hyperspectral features between 400–1350 nm combined with machine learning (ML) methods in the individual and whole growth stages. In this study, 3 years of winter wheat plot experiments were conducted. Ground-level canopy hyperspectral reflectance and corresponding plant NBI values were measured during the jointing, booting, flowering and filling stages. Continuous removal spectra (CRS) and logarithmic transformation spectra (LOGS) were derived from the original canopy spectra. Sensitive bands and vegetation indices (VIs) highly correlated with NBI under different spectral transformations were selected as hyperspectral features to construct the NBI estimation models combined with ML algorithms. The study indicated that the spectral transformation significantly improved the correlation between the sensitive bands, VIs and the NBI. The correlation coefficient of the sensitive band in CRS in the booting stage increased by 27.87%, reaching −0.78. The leaf chlorophyll index (LCI) in LOGS had the highest correlation with NBI in the filling stage, reaching a correlation coefficient of −0.96. The NBI prediction accuracies based on the sensitive band combined with VIs were generally better than those based on the univariate hyperspectral feature, and the prediction accuracy of each growth stage was better than that of the whole growth stage. The random forest regression (RFR) method performed better than the support vector regression (SVR) and partial least squares regression (PLS) methods. The NBI estimation model based on the LOGS-RFR method in the filling stage could explain 95% of the NBI variability with relative prediction deviation (RPD) being 3.69. These results will provide a scientific basis for better nitrogen nutrition monitoring, diagnosis, and later for field management of winter wheat.

Estimation of Winter Wheat Canopy Chlorophyll Content Based on Canopy Spectral Transformation and Machine Learning Method

Abstract: Canopy chlorophyll content (CCC) is closely related to crop nitrogen status, crop growth and productivity, detection of diseases and pests, and final yield. Thus, accurate monitoring of chlorophyll content in crops is of great significance for decision support in precision agriculture. In this study, winter wheat in the Guanzhong Plain area of the Shaanxi Province, China, was selected as the research subject to explore the feasibility of canopy spectral transformation (CST) combined with a machine learning method to estimate CCC. A hyperspectral canopy ground dataset in situ was measured to construct CCC prediction models for winter wheat over three growth seasons from 2014 to 2017. Sensitive-band reflectance (SR) and narrow-band spectral index (NSI) were established based on the original spectrum (OS) and CSTs, including the first derivative spectrum (FDS) and continuum removal spectrum (CRS). Winter wheat CCC estimation models were constructed using univariate regression, partial least squares (PLS) regression, and random forest (RF) regression based on SR and NSI. The results demonstrated the reliability of CST combined with the machine learning method to estimate winter wheat CCC. First, compared with OS-SR (683 nm), FDS-SR (630 nm) and CRS-SR (699 nm) had a larger correlation coefficient between canopy reflectance and CCC; secondly, among the parametric regression methods, the univariate regression method with CRS-NDSI as the independent variable achieved satisfactory results in estimating the CCC of winter wheat; thirdly, as a machine learning regression method, RF regression combined with multiple independent variables had the best winter wheat CCC estimation accuracy (the determination coefficient of the validation set (Rv2) was 0.88, the RMSE of the validation set (RMSEv) was 3.35 and relative prediction deviation (RPD) was 2.88). Thus, this modeling method could be used as a basic method to predict the CCC of winter wheat in the Guanzhong Plain area.

Identification and Classification of Maize Drought Stress Using Deep Convolutional Neural Network

Abstract: Drought stress seriously affects crop growth, development, and grain production. Existing machine learning methods have achieved great progress in drought stress detection and diagnosis. However, such methods are based on a hand-crafted feature extraction process, and the accuracy has much room to improve. In this paper, we propose the use of a deep convolutional neural network (DCNN) to identify and classify maize drought stress. Field drought stress experiments were conducted in 2014. The experiment was divided into three treatments: optimum moisture, light drought, and moderate drought stress. Maize images were obtained every two hours throughout the whole day by digital cameras. In order to compare the accuracy of DCNN, a comparative experiment was conducted using traditional machine learning on the same dataset. The experimental results demonstrated an impressive performance of the proposed method. For the total dataset, the accuracy of the identification and classification of drought stress was 98.14% and 95.95%, respectively. High accuracy was also achieved on the sub-datasets of the seedling and jointing stages. The identification and classification accuracy levels of the color images were higher than those of the gray images. Furthermore, the comparison experiments on the same dataset demonstrated that DCNN achieved a better performance than the traditional machine learning method (Gradient Boosting Decision Tree GBDT). Overall, our proposed deep learning-based approach is a very promising method for field maize drought identification and classification based on digital images.

Cloud-Fog based framework for drought prediction and forecasting using artificial neural network and genetic algorithm

Abstract: Drought is one of the most recurrent natural disasters with cataclysmic effects on water budget, crop production, economic progression and public health. These consequences are magnified by the cli...

Early Prediction of Soybean Traits through Color and Texture Features of Canopy RGB Imagery

Abstract: Global crop production is facing the challenge of a high projected demand, while the yields of major crops are not increasing at sufficient speeds. Crop breeding is an important way to boost crop productivity, however its improvement rate is partially hindered by the long crop generation cycles. If end-season crop traits such as yield can be predicted through early-season phenotypic measurements, crop selection can potentially be made before a full crop generation cycle finishes. This study explored the possibility of predicting soybean end-season traits through the color and texture features of early-season canopy images. Six thousand three hundred and eighty-three images were captured at V4/V5 growth stage over 6039 soybean plots growing at four locations. One hundred and forty color features and 315 gray-level co-occurrence matrix-based texture features were derived from each image. Another two variables were also introduced to account for location and timing differences between the images. Five regression and five classification techniques were explored. Best results were obtained using all 457 predictor variables, with Cubist as the regression technique and Random Forests as the classification technique. Yield (RMSE = 9.82, R2 = 0.68), Maturity (RMSE = 3.70, R2 = 0.76) and Seed Size (RMSE = 1.63, R2 = 0.53) were identified as potential soybean traits that might be early predictable.