Assessment of wheat chlorophyll content by the multiple linear regression of leaf image features

The measurement of crop nutrition is considerably significant in agricultural practices,especially in the application of mechanized variable rate fertilization.Feature extraction and model building are two important links in crop nutrition measurement by digital image.In this paper,a feature set of fusion multi-colour space in field prototype is extracted and an evaluation approach using stepwise-based ridge regression(SBRR)that uses correlation-based evaluation method is employed.First the image features of three known colour spaces are extracted,meanwhile a new colour space named rgb is constructed according to the characteristics that RGB colour space easily affected by light.Then the SBRR with nested cross validation is used to find the best evaluation model.By performance evaluation,the optimal SBRR model is obtained(R^(2)=0.718 RMSE=5.111).Additionally,compared with two other nutritional evaluation approach named backpropagation artificial neural network(BP-ANN)and k-nearest neighbors(KNN),SBRR achieves better performance in both R^(2) and RMSE.Furthermore the proposed model’s reliability is verified using the image dataset taken from the same wheat field in the next year.The R^(2) and RMSE values are 0.794 and 4.304,respectively.The comparisons and verification show that our proposed SBRR approach can achieve better experimental results and can be considered a reliable and low-cost alternative for estimating the chlorophyll content of wheat leaves in field.

Passive-Event-Assisted Approach for the Localizability of Large-Scale Randomly Deployed Wireless Sensor Network

Localizability in large-scale, randomly deployed Wireless Sensor Networks(WSNs) is a classic but challenging issue. To become localizable, WSNs normally require extensive adjustments or additional mobile nodes. To address this issue, we utilize occasional passive events to ease the burden of localization-oriented network adjustment. We prove the sufficient condition for node and network localizability and design corresponding algorithms to minimize the number of nodes for adjustment. The upper bound of the number of adjusted nodes is limited to the number of articulation nodes in a connected graph. The results of extensive simulations show that our approach greatly reduces the cost required for network adjustment and can thus provide better support for the localization of large-scale sparse networks than other approaches.