Visible-near-infrared spectrum (Vis-NIR) spectroscopy technology is one of the most important methods for non-destructive and rapid detection of soil total nitrogen (STN) content. In order to find a practical way to build STN content prediction model, three conventional machine learning methods and one deep learning approach are investigated and their predictive performances are compared and analyzed by using a public dataset called LUCAS Soil (19,019 samples). The three conventional machine learning methods include ordinary least square estimation (OLSE), random forest (RF), and extreme learning machine (ELM), while for the deep learning method, three different structures of convolutional neural network (CNN) incorporated Inception module are constructed and investigated. In order to clarify effectiveness of different pre-treatments on predicting STN content, the three conventional machine learning methods are combined with four pre-processing approaches (including baseline correction, smoothing, dimensional reduction, and feature selection) are investigated, compared, and analyzed. The results indicate that the baseline-corrected and smoothed ELM model reaches practical precision (coefficient of determination (R²) = 0.89, root mean square error of prediction (RMSEP) = 1.60 g/kg, and residual prediction deviation (RPD) = 2.34). While among three different structured CNN models, the one with more 1 × 1 convolutions preforms better (R² = 0.93; RMSEP = 0.95 g/kg; and RPD = 3.85 in optimal case). In addition, in order to evaluate the influence of data set characteristics on the model, the LUCAS data set was divided into different data subsets according to dataset size, organic carbon (OC) content and countries, and the results show that the deep learning method is more effective and practical than conventional machine learning methods and, on the premise of enough data samples, it can be used to build a robust STN content prediction model with high accuracy for the same type of soil with similar agricultural treatment.

Agricultural robots are flexible to obtain ambient information across large areas of farmland. However, it needs to face two major challenges: data compression and filtering noise. To address these challenges, an encoder for ambient data compression, named Tiny-Encoder, was presented to compress and filter raw ambient information, which can be applied to agricultural robots. Tiny-Encoder is based on the operation of convolutions and pooling, and it has a small number of layers and filters. With the aim of evaluating the performance of Tiny-Encoder, different three types of ambient information (including temperature, humidity, and light) were selected to show the performance of compressing raw data and filtering noise. In the task of compressing raw data, Tiny-Encoder obtained higher accuracy (less than the maximum error of sensors ±0.5°C or ±3.5% RH) and more appropriate size (the largest size is 205 KB) than the other two auto-encoders based convolutional operations with different compressed features (including 20, 60, and 200 features). As for filtering noise, Tiny-Encoder has comparable performance with three conventional filtering approaches (including median filtering, Gaussian filtering, and Savitzky-Golay filtering). With large kernel size (i.e., 5), Tiny-Encoder has the best performance among these four filtering approaches: the coefficients of variation with the large kernel (i.e., 5) were 8.6189% (temperature), 10.2684% (humidity), 57.3576% (light), respectively. Overall, Tiny-Encoder can be used for ambient information compression applied to microcontrollers in agricultural information acquisition robots. Keywords: agricultural information, robot, ambient information, data compression, embedded machine learning methods DOI: 10.25165/j.ijabe.20221504.6911 Citation: Wang Y T, Li M Z, Ji R H, Wang M J, Zhang Y, Zheng L H. Novel encoder for ambient data compression applied to microcontrollers in agricultural robots. Int J Agric & Biol Eng, 2022; 15(4): 197–204.