Research in Agricultural Engineering - In Press
An effective machine learning model for estimation of reference evapotranspiration under data limited conditions Original Paper
Saravanan Karuppanan (orcid: 0000-0002-1971-5239), Saravanan Ramasamy, Balaji Lakshminarayanan, Sreemanthrarupini Nariangadu Anuthaman
Reference crop evapotranspiration (ETo) is a vital hydrological component influenced by various climate variables that impact water and energy balances. It plays a crucial role in determining crop water requirements and irrigation scheduling. Despite the availability of numerous approaches for estimation, accurate and reliable ETo estimation is essential for effective irrigation water management. Therefore, this study aimed to identify the most suitable machine learning model for assessing ETo using observed daily values of limited input parameters in tropical savannah climate regions. Three machine learning models—long short-term memory (LSTM) neural network , artificial neural network (ANN), and support vector regression (SVM)—were developed with four different input combinations, and their performances were compared with those of locally calibrated empirical equations. The models were evaluated using statistical indicators such as the root mean square error (RMSE), coefficient of determination (R2), and Nash-Sutcliffe efficiency (NSE). The results showed that the LSTM model, using the combination of temperature and wind speed, provided more reliable predictions with R2 values greater than 0.75 and RMSEs less than 0.63 mm/day across all the considered weather stations. This study concludes that, especially under limited data conditions, the developed deep learning model improves ETo estimation more accurately than empirical models for tropical climatic regions.
Anaerobic bio-processing of agricultural wastes for biotechnological production of lactic acid and volatile fatty acid by landfill soil inoculumsOriginal Paper
Darwin Darwin
With the growing global population and the ensuing surge in organic waste, effective management strategies are crucial to prevent environmental pollution. This study aims to address this challenge by exploring the utilization of organic waste (OW) as a substrate for the production of lactic acid (LA) and volatile fatty acids (VFA) through anaerobic digestion (AD) bioprocessing. This study explores natural anaerobic bioprocessing, employing OW substrates (elephant grass/G, cassava waste/S, and fruit waste/F) with non-sterile inoculum landfill soil (LS) from the Banda Aceh City landfill Site. AD is conducted by varying the concentration of substrate/inoculum, with concentrations set at 50, 100, and 150 g/L. Including LA and VFA production stands out as a promising avenue. The results unveil that the digester, fuelled with 150 g/L of F substrate, exhibits the highest concentration of LA, reaching an impressive 25 mmol/L. Furthermore, noteworthy is the observation that the digester, fuelled with 100 g/L S substrate, also manifests a significant LA production (18.50 mmol/L). Similarly, the digester, supplied with 150 g/L cassava waste, showcases the highest VFA concentration, a remarkable 92.5 mmol/L. Intriguingly, the anaerobic bioprocessing of G substrate did not produce LA, yet all substrates showcased VFA production, albeit with fluctuating and lower concentrations. This study highlights the potential of incorporating Simply Sugar for enhanced LA production and starch-based substrates for increased VFA production when utilizing LS as the inoculum. The conducted bioprocessing process presents promising and groundbreaking outcomes for future development in sustainable waste utilization.
Use of Thermal Imaging Camera for Wild Animal Detection along RoadsOriginal Paper
Jiří Brožovský, Veronika Hartová, Martin Kotek, Jan Hart, Jitka Kumhálová
Vehicle collisions with wild animals are a common problem on roads, with significant impacts on road safety and wildlife populations. Collisions with wild animals are one of the most frequent road accidents. According to police statistics, there were nearly 16,000 road accidents caused by collisions with wild animals in the Czech Republic in 2019. Collisions with deer are the most common. There are several technologies and measures that can help reduce the risk of a vehicle colliding with a wild animal. One of the technologies used is a night vision system based on infrared spectrum sensing. This technology is slowly becoming part of the equipment of, in particular, premium car brands due to its high cost. This paper tested a low-cost solution using a commercially available thermal imaging camera and found a substantial reduction in the time to detect wild animals along the road, namely in the order of seconds.
Perception of bimodal warning cues during remote supervision of autonomous agricultural machinesOriginal Paper
ANITA CHIDERA EZEAGBA, CHERYL MARY GLAZEBROOK, DANNY DELMAR MANN
Agricultural machines that are fully autonomous will still need human supervisors to monitor and troubleshoot system failures. Recognizing the emergency as soon as possible is crucial in these circumstances to reduce adverse effects. The ability of humans to detect visual, auditory, or tactile cues is usually enabled by warning systems. The effectiveness of different warning cues varies in terms of prompting a quick response. The objective of the study was to compare the effectiveness of two bimodal warnings (i.e., visual-auditory and visual-tactile) at eliciting supervisor perception (which equates to level one situation awareness). Twenty-five participants engaged with an autonomous sprayer simulation. Two realistic remote supervision scenarios (i.e., in-field and close-to-field) were used to examine two bimodal warning cues: i) visual-auditory and ii) visual-tactile. Effectiveness of each bimodal warning was assessed based on two measures: i) response time and ii) noticeability. There was no significant difference between the bimodal warning cues in terms of response time when tractor sound was present in the experimental environment (reflecting the in-field remote supervision scenario), however visual-tactile cues yielded shorter response times than visual-auditory cues when the experimental environment was quiet (reflecting the close-to-field remote supervision scenario. There were no statistically significant differences between visual-auditory and visual-tactile warnings with respect to noticeability. Participants' subjective answers indicated that they preferred the visual-tactile cues better than the visual-auditory cues. It is concluded that visual-tactile warnings are preferred over visual-auditory warnings to enable perception during remote supervision of autonomous agricultural machines (AAMs).
Towards Interpretability: Assessment of Residual Networks for Tomato Leaf Disease ClassificationOriginal Paper
Raphael Berdin, Rob Christian Caduyac
Tomato occupies a prominent place in Philippines’ agricultural economy. However, tomato leaf diseases are challenges in tomato crop production leading to economic losses. Among the tomato leaf diseases, early blight and Septoria leaf spot are prevalent in the Philippines due to the climate. Thus, accurate identification of diseases affecting tomato leaves is essential. Currently, visual inspection is the primary method for diagnosing tomato leaf diseases which is time consuming and inefficient. This study aims to develop a quantized Residual Network with convolutional 50 layers (ResNet-50) based model to classify tomato leaves as healthy or affected by Septoria leaf spot or early blight. Furthermore, to enhance the reliability of the models' classification, Grad-CAM was implemented. In contrast with visual inspection, a programmed system does not get tired and can provide consistent performance. As a result, the original 32-bit floating point model attained an accuracy rate of 91.22%. The quantized 16-bit floating point model demonstrated comparable performance with 90.10% accuracy with 50% reduction in model size and inference time of 0.3942 seconds. The minimal accuracy loss of the 16-bit model relative to the 32-bit model is due to post-training quantization. The reduction to 16-bit precision is significant for future edge devices deployment where resources are limited.
Spoilage Detection of Tomatoes using Convolutional Neural NetworkOriginal Paper
Ninja Begum, Hazarika Manuj Kumar
With the increasing productivity in agriculture, it has become very much essential to look for an advanced technique that will help in minimizing losses. Recently deep learning has outperformed the task of recognition and classification of fruits and vegetables automatically from images, finding applicability in this study. This work thus attempts to develop an automatic spoilage detection CNN model for tomatoes. In this work, a deep learning based CNN model is trained and validated on a self-prepared dataset for classifying tomatoes as edible and spoilt is proposed. The dataset consisted of 810 images, out of which 572 images are considered for training and 238 images for validation. The model is also trained iteratively with varying epoch and batch sizes to evaluate the model in giving the highest accuracy in classification. The highest accuracy of 99.70% is achieved at epoch 20 and batch size 32. Further evaluating the performance of the developed model using confusion matrix, precision, recall and accuracy of 100%, 87% and 95% respectively was obtained for spoilage detection of tomatoes. Also on establishing Pearsons’ correlation between the predictive model and the sensory evaluation results, a Pearson correlation of 0.895, showing that there is strong linear correlation between them.
Development of smart microirrigation system using ARDUINO UNO for okra cultivation in BangladeshOriginal Paper
Sharmin Akter, Md Mostafizar Rahman, Rafatul Zannat, Md Masud Rana, Md Moinul Hossain Oliver, Md Aslam Ali
Conventional irrigation practices result in a substantial amount of water loss for Okra cultivation. Although microirrigation can address this issue by delivering water directly near the rootzone, it requires manual operation. These issues, however, can be resolved with the introduction of a smart microirrigation system. This study aims to develop a smart microirrigation system for okra, in conjunction with the sub-components of drip irrigation, a microcontroller, and a soil sensor. The experiment was laid out with a randomized complete block design (RCBD) having three treatments: Control Irrigation (T1), Drip Irrigation (T2), and Smart Microirrigation (T3). The experimental field was irrigated based on soil moisture regimes in the crop rootzone. Plant growth, yield, and water use efficiency were assessed to evaluate the system. The results showed no significant differences among these treatments (at P < 0.05). The best water usage efficiency (15.98 kg·m-3 ) was observed in the T3 treatment, which also provided about 13.10% water savings compared to conventional irrigation. This study indicates that smart microirrigation system could be a promising technology for water-efficient okra cultivation.