Time–frequency Analysis for Biosystems Engineering

Abstract Many biological and agricultural systems naturally emit a number of signals which may be easily gathered and then used to monitor or control the system. For example, signals consisting of measurements of the temperature or heart rate of animals may be used to assess the health of the individuals. The interpretation of these signals is relatively simple since the value at any time gives some indication of the animal's condition. Other signals such as acoustic or vibration measurements are less easy to interpret. Here it is the spectral content, that is the amplitude and rate of oscillations within the signal, that is important rather than the actual measurement value or trend. Often such signals are complicated and their spectral content varies with time. This paper provides a review of methods, known as time–frequency analyses, that accurately track these variations. Researchers in other areas have found that traditional time–frequency techniques such as the short-time Fourier transform are unable to calculate the spectral content of highly transient signals with sufficient accuracy. For this reason new signal processing methods such as wavelets and energy distributions have been developed. In this review these methods are described and their application to the output from biological or agricultural systems is discussed. Six different analysis techniques are detailed. Each has its own advantages and disadvantages and is suited to different types of signal. The aim is to give objective criteria that may be used to choose the correct analysis technique for a particular signal. To this end, key properties of each method are compared, such as the precision or resolution, the form and ease of interpretation of results, the ability to separate the relevant parts of the signal from noise, the complexity of computation required and the pitfalls or artefacts that may occur. The motivation for this review is to enable researchers to select the correct tools required to investigate the information content of previously unstudied signals. Information from such signals may be of great value in a number of biosystems engineering applications such as automatic control of farm machinery or integrated monitoring systems for livestock.