We present a decision support system for managing water quality in prawn ponds. The system uses various sources of data and deep learning models in a novel way to provide 24-h forecasting and anomaly detection of water quality parameters. It provides prawn farmers with tools to proactively avoid a poor growing environment, thereby optimising growth and reducing the risk of losing stock. This is a major shift for farmers who are forced to manage ponds by reactively correcting poor water quality conditions. To our knowledge, we are the first to apply Transformer as an anomaly detection model, and the first to apply anomaly detection in general to this aquaculture problem. Our technical contributions include adapting ForecastNet for multivariate data and adapting Transformer and the Attention model to incorporate weather forecast data into their decoders. We attain an average mean absolute percentage error of 12% for dissolved oxygen forecasts and we demonstrate two anomaly detection case studies. The system is successfully running in its second year of deployment on a commercial prawn farm.

The research presented in this paper was conducted as part of a project that aimed at using biosensors on sentinel oysters to provide a biological perspective of environmental changes. The physiological response patterns (e.g. heart rate variability, shell gape) of sentinel oysters can provide valuable insight into their exposure to environmental stressors. Sensors were attached to measure both heart rate and shell gape (physiological behaviour) of oysters placed at different depths in the water column. Sensors were also deployed to measure water quality that represents the environmental condition the oysters were in. The objective of this study is to utilise the data from different sensors to investigate how environmental conditions modulate physiological response of oysters. We have utilised a set of machine learning models to develop data-driven approaches that can predict heart rate and shell gape (opening and closing action) of oysters from water quality variables, and vice-versa. The level of prediction accuracy indicates how well environmental conditions influence the physiological response of oysters. The effectiveness of the developed approaches is evaluated using data collected from two different deployments of sensors in South East Tasmania, Australia. Experimental results demonstrate that the presented data-driven approaches can provide accurate predictions of physiological and water quality variables, for the data set considered in this study. The prediction error (in terms of MAPE) for HR and water quality is in the range of 3.17%-8.21% and 0.72%-2.48%, respectively, and classification accuracy (F-Score) for shell gape varies between 0.96 and 0.99.

Many hybrids of marine molluscs show improved growth in comparison to their pure parental species. Yet, little is known about the physiological mechanisms underlying the better hybrid performance. In this study, movement, oxygen consumption rate (ṀO2), and heart rate were determined in 22month old cultured abalone Haliotis rubra, H. laevigata and their interspecies hybrid, the latter of which exhibits improved growth rate. Abalone were exposed to an acute temperature increase following acclimation to 16 or 23°C at high and low oxygen levels (100% or 70% air saturation, respectively). Movement of hybrids and H. laevigata was generally not affected by temperature and oxygen levels, yet H. rubra showed a strong thermal response. Heart rate and ṀO2/temperature slopes revealed that hybrids were least affected by oxygen levels. Arrhenius break-point temperatures of hybrids and H. laevigata, but not H. rubra, were generally higher when abalone were acclimated to 23°C in comparison to 16°C. The hybrid had more stable maximum heart rate and ṀO2 values across acclimation conditions in comparison to H. laevigata and H. rubra. Thus, it appears that hybrids are able to maintain physiological functions over a broader environmental range. This improved tolerance to environmental fluctuations may bolster energy metabolism and improve growth in variable environments such as aquaculture farms.

The problem of learning long-term dependencies in sequences using Recurrent Neural Networks (RNNs) is still a major challenge. Recent methods have been suggested to solve this problem by constraining the transition matrix to be unitary during training which ensures that its norm is equal to one and prevents exploding gradients. These methods either have limited expressiveness or scale poorly with the size of the network when compared with the simple RNN case, especially when using stochastic gradient descent with a small mini-batch size. Our contributions are as follows; we first show that constraining the transition matrix to be unitary is a special case of an orthogonal constraint. Then we present a new parametrisation of the transition matrix which allows efficient training of an RNN while ensuring that the matrix is always orthogonal. Our results show that the orthogonal constraint on the transition matrix applied through our parametrisation gives similar benefits to the unitary constraint, without the time complexity limitations.

An experiment is introduced which demonstrates the application of supervised feature learning using a Convolutional Neural Network for cattle behaviour classification. The data set used contains observations from sensors attached to the cattle. Previously this problem was addressed by classifying features learned by a stacked autoencoder. Here we explore an alternative method for learning effective features. Convolutional Neural Networks have shown immense success in computer vision, natural language processing, speech recognition etc. The success of Convolutional Neural Network in so many applications has inspired us to verify how effective this network is on learning features from cattle data. A shallow Convolutional Neural Network such as we have developed in this experiment learns effective features for classification and is time-efficient compared to previous model.

Accurate models for one day ahead prediction of stream flow are crucial for water management catchment scale for agriculture. This is particularly important for a country such as Australia where weather conditions can be harsh and varying. The Support Vector Regression (SVR) and the Vector Auto Regression (VAR) are standard methods used for time series prediction [12, 14]. However, since they use fixed-sized time windows, these models cannot capture long-term dependencies that are often present in stream flow time series. Recurrent Neural Networks (RNNs) do not have this weakness, yet they have not been extensively used on this type of time series. In this work, we tested various types of RNN architectures, including the recently introduced clock-work RNN (CW-RNN) [10], on two different stream flow datasets in Tasmania. We compared their accuracy with that of the SVR and VAR methods on the task of one day ahead prediction of the stream flow. In our experiments, the CW-RNNs outperformed the SVR and VAR methods across both datasets. In particular, when evaluated on the largest test set, which contained approximately 4 years of daily records, the normalised Root Mean Squared Error (nRMSE) and the Nash-Sutcliffe Efficiency (NSE) of the best CW-RNN architecture were equal to 0.166 and 0.956, respectively. This is a significant improvement over the best SVR model, which had nRMSE = 0.202 and NSE = 0.936. Our results suggest that RNNs are well suited for the task of one day ahead prediction of stream flow.

For remote characterization of inaccessible underground mine voids, we are developing unmanned aerial vehicles (equipped with multiple sensors, including cameras) to fly into the mine voids to map their shape, condition, and most importantly, mineralization of the surface. The X-ray fluorescence (XRF) spectroscopy analysis is normally conducted on rock samples in order to detect the present elements (that constitutes minerals). Mining company staffs, however, are able to judge rock types based upon visual features alone. This implies that there are some associations between the XRF signatures and the visual features of rocks. Inspired by this, we have developed a machine learning approach to predict the presence of elements in rocks, for inferring probable rock and mineral types, from imaging features. Note that there exist a number of works in the literature for classifying rocks from digital images. However, to the best of our knowledge, limited attempt has been made to find association between the digital imaging features and the XRF signatures for mineralogy discovery that we have addressed in this paper. The machine learning algorithm is trained offline based on visual imaging and XRF spectroscopy analysis data of collected rock samples in a laboratory. The imaging features provide the visual cues, and the XRF data provide information on element presence/concentration. The machine learning algorithm (regression) discovered the non-linear relationship between these feature spaces and was able to predict the element presence with high accuracy as evidenced from the experimental results.

An L-band 19-element multibeam feed has been designed for the proposed Five-hundred-meter Aperture Spherical radio Telescope. Physical optics analysis of the full reflector with mutlibeam feed has been undertaken. The analysis results are presented here.

An in situ optical oyster heart rate sensor generates signals requiring frequency estimation with properties different to human ECG and speech signals. We discuss the method of signal generation and highlight a number of these signal properties. An optimal heart rate estimation approach was identified by application of a variety of frequency estimation techniques and comparing results to manually acquired values. Although a machine learning approach achieved the best performance, accurately estimating 96.8% of the heart rates correctly, a median filtered autocorrelation approach achieved 93.7% with significantly less computational requirement. A method for estimating heart rate variation is also presented.

In this paper, two new multivariate time series classifiers are introduced as the Bag of Class Posteriors (BOCP) and the Bag of Class Posterior with Ordering (BOCPO). The models propose a new multi-scale feature representation where the class posterior estimates of contiguous local patterns are aggregated over longer time scales. The models are employed as part of an animal behaviour monitoring system that are comprised of sensors, which are fitted to the animals, and a classifier that translates sensor data into knowledge of the animal's behaviour.Animal monitoring systems are commonly developed to infer a small number of behaviours with relevance to a specific application. To investigate if a standard monitoring system with an Inertial Measurement Unit (IMU) can be reused for different management applications, a set of ten cattle behaviours relevant to different management applications were classified with the proposed models. Results indicate that the multi-scale BOCP and BOCPO models were far more capable of classifying the cow behaviours offering a 43% to 77% improvement over benchmark time interval classifiers with fixed time resolution. In addition, the BOCPO model was shown to offer a far more efficient feature representation than the related multi-scale Bag of Features (BOF) classifier (up to 200 times smaller) making it better suited to deploy upon monitoring devices fitted to animals in the field. Crown Copyright (C) 2014 Published by Elsevier Ltd. All rights reserved.

BACKGROUND:Despite ongoing reduction in genotyping costs, genomic studies involving large numbers of species with low economic value (such as Black Tiger prawns) remain cost prohibitive. In this scenario DNA pooling is an attractive option to reduce genotyping costs. However, genotyping of pooled samples comprising DNA from many individuals is challenging due to the presence of errors that exceed the allele frequency quantisation size and therefore cannot be simply corrected by clustering techniques. The solution to the calibration problem is a correction to the allele frequency to mitigate errors incurred in the measurement process. We highlight the limitations of the existing calibration solutions such as the fact they impose assumptions on the variation between allele frequencies 0, 0.5, and 1.0, and address a limited set of error types. We propose a novel machine learning method to address the limitations identified.RESULTS:The approach is tested on SNPs genotyped with the Sequenom iPLEX platform and compared to existing state of the art calibration methods. The new method is capable of reducing the mean square error in allele frequency to half that achievable with existing approaches. Furthermore for the first time we demonstrate the importance of carefully considering the choice of training data when using calibration approaches built from pooled data.CONCLUSION:This paper demonstrates that improvements in pooled allele frequency estimates result if the genotyping platform is characterised at allele frequencies other than the homozygous and heterozygous cases. Techniques capable of incorporating such information are described along with aspects of implementation.

Background: The costs associated with developing high density microarray technologies are prohibitive for genotyping animals when there is low economic value associated with a single animal (e.g. prawns). DNA pooling is an attempt to address this issue by combining multiple DNA samples prior to genotyping. Instead of genotyping the DNA samples of the individuals, a mixture of DNA samples (i.e. the pool) from the individuals is genotyped only once. This greatly reduces the cost of genotyping. Pooled samples are subject to greater genotyping inaccuracies than individual samples. Wrong genotyping will lead to wrong biological conclusions. It is thus required to calibrate the resulting genotypes (allele frequencies).Methods: We present a regression based approach to translate raw array output to allele frequency. During training, few pools and the individuals that constitute the pools are genotyped. Given the genotypes of individuals that constitute the pool, we compute the true allele frequency. We then train a regression algorithm to produce a mapping between the raw array outputs to the true allele frequency. We test the algorithm using pool samples withheld from the training set. During prediction, we use this map to genotype pools with no prior knowledge of the individuals constituting the pools.Results and discussion: After data quality control we have available a dataset comprised of 912 pools. We estimate allele frequency using three approaches: the raw data, a commonly used piecewise linear transformation, and the proposed local-global learner fusion method. The resulting RMS errors for the three approaches are 0.135, 0.120, and 0.080 respectively. Crown Copyright (C) 2015 Published by Elsevier Ltd. All rights reserved.

This study investigated the applicability of machine learning algorithms to detect the presence of elements in underground mines from rock surface images, which is proposed as a heuristic classification method inspired by the ability of human geologists to make judgments about the location of ore veins by eye. A regression algorithm was investigated to find associations between image features and X-Ray Fluorescence (XRF) signatures indicating elemental content of the surface and near-surface region of the rocks. A set of image processing algorithms was used to extract color distribution, edge orientation statistics, and texture of the rock surfaces. XRF signatures were obtained from the same samples, providing a semi-quantitative measure of element concentration. The process was performed on a set of 20 rock samples. The regression algorithm was then trained to find a mapping between image features and the semi-quantitative element concentrations (corresponding with XRF peaks). Experimental results demonstrate the potential effectiveness of the proposed approach in the context of a specific ore body.

Models were developed to classify six different behaviours for a group of seven steers fitted with an accelerometer and pressure sensor. As part of the process, a greedy feature selection method was used to identify the most discriminatory inputs from a diverse set of statistical, spectral and information theory based features. The study showed the second order statistic features (standard deviation and sum of absolute values), which represent the level of motion intensity, were the most discriminatory individual features. The classification performance of models were further enhanced by using spectral features (with statistical features) to capture the periodicity of head movements and to differentiate between the dominant frequencies of various motions. Incorporating feature selection into model development not only improves model performance, but assists in understanding the different motion characteristics that enable behaviours to be discriminated.

Due to frost and insufficient exposure to sunlight, some grape bunches remain undeveloped during harvesting. For automation of harvesting, it is required to automatically identify the mature grape bunches. This paper presents a sequence of image processing and computational intelligence methods to identify mature grape bunches. It's a two-step process where in the first step the grape bunches are separated from the background of an image and in the second step the grape bunch is classified into mature and undeveloped group. We achieved 96.88% accuracy on the images obtained from a strip of vineyard in Cambridge, Tasmania.

The properties of a unit-fraction section of a spherical Luneburg lens, with two backing ground planes intersecting at the spherical lens's axis, have been evaluated and compared to experimental results. Compared to a full Luneburg lens, the advantage of using a fraction of a Luneburg lens is the reduced weight of the antenna. This reduction is achieved at the expense of the scanning range of the antenna. An analytical expression relating the size of the ground plane to the scanning range is derived. The measured radiation patterns, gain, and scan loss of a hemispherical and of a quarter-sphere lens antenna are presented. The effects of the ground planes on the S parameters of the feed antenna are discussed, and simulated results are compared with measured results.

The combination of low density SNP arrays and DNA pooling is a fast and cost effective approach to genotyping that opens up basic genomics to a range of new applications and studies. However we have identified significant limitations in the existing approach to calculating allele frequencies with DNA pooling. These limitations include a reduced ability to deal with SNP to SNP variation via the standard interpolation method. Our contribution is a new hierarchical learning framework which resolves these drawbacks. The framework involves a hierarchy of two greedily trained layers of learners. The first layer learns the bias of each SNP then applies a calibration to reduce SNP bias by mapping into a common coordinate system across all SNPs. The second layer learns an allele frequency function exploiting the global SNP data. A range of algorithms have been applied including linear regression, neural network and support vector regression. The framework has been tested on pooled samples of Black Tiger prawns that have been genotyped with low density Sequenom iPLEX panels. Analysis of pooled samples and the corresponding individually genotyped SNP samples indicate the pooling approach introduces an allele frequency RMS error of 0.12. The existing calibration approach corrects ~14% of the error. Our hierarchical approach is 4.5 times as effective by correcting for ~64% of the introduced error. This is a significant reduction and has the potential to enable genetic studies previously not possible due to allele frequency error. Although testing so far is limited to low density SNP arrays the approach was developed to generalize to other SNP genotyping technologies.

New sensor streams are being generated at a rapidly increasing rate. The sources of these streams are a diverse set of networked sensors, diverse both in sensing hardware and sensing modality. Machine learning algorithms are ideally placed to develop generalized methods for stream analysis. One exemplar problem is the detection and analysis of periodic structure within these streams. Our contribution is the proposal of a new machine learning framework that (i) classifies a signal as periodic or aperiodic, (ii) further analyses the signal to find periodic structure using a neural network, and (iii) groups the motifs in the periodic signals using a modified Self Organising Map algorithm. We also demonstrate the framework using data generated by an Oyster heart rate sensor. We find that the generalized approach our classifier improves the detection of signal periods by reducing the number of functions classified as periodic from 11% to 9%; however, most benefit occurs for period calculation with the number of erroneously calculated periods reducing from 14% to 4%.

An experiment to study the impact of supplements upon the feeding behavior of dairy cattle was conducted at the Tasmanian Institute of Agriculture (TIA) Dairy Research Facility. Collar systems with 3-axis accelerometer and magnetometer were fitted to individual cows to infer their feeding behavior. We describe the solutions applied to correct for sensor data issues, and then provide some preliminary analysis associated with developing behavior models using multivariate time series data.

We present the design and implementation of a high-temperature superconducting (HTS) Josephson junction detector array for terahertz (THz) imaging. The array device is made of ten YBa2Cu3Ox−7 (YBCO) step-edge junctions coupled to gold thin-film ring-slot antennas on a MgO substrate. The design and characterization of the detector array in response to a 0.6 THz signal are presented. The development of multi-channel biasing and read-out electronics and the system integration with a commercial cryocooler are also described.