This book is written for three overlapping audiences who all face the same problem from different sides: how to turn raw physical measurements into reliable machine perception. We built it so that one volume serves all three without forcing anyone to read at the wrong depth.
Three audiences, one book
The first audience is advanced undergraduates in engineering, computer science, and applied physics who have met sensors in a lab but never systematically. For you the book is a first coherent bridge from a temperature reading or an accelerometer trace to a trained model that acts on it.
The second audience is graduate students and instructors. If you are building a thesis around inertial navigation, biosignal classification, or multimodal fusion, the later chapters give you the estimators, the noise models, and the deep architectures used in current practice. Instructors can lift Part I as a self-contained foundations module and assemble a course from the level tags described below.
The third audience is practicing researchers and engineers shipping real systems: IoT and wearables, robotics and autonomous vehicles, industrial condition monitoring, medical devices, and environmental sensing. You will find the calibration, drift, and uncertainty material that textbooks usually omit and that production systems live or die by.
Find your depth: the level tags
Every chapter and every section carries one of four tags. The tag names the intended altitude, so you can skim, study, or skip with intent. Reading only the [F] and [C] material gives a complete, honest first pass through the whole field; the [A] and [R] tags mark where you go deeper once the core is solid.
| Tag | Level | Who it is for |
|---|---|---|
| [F] | Foundational | First exposure; assumes only the prerequisites below. |
| [C] | Core | The working knowledge every sensory-AI practitioner needs. |
| [A] | Advanced | Specialist depth for a chosen modality or method. |
| [R] | Research frontier | Open problems and recent results at the edge of the field. |
What we assume, and what we teach
The prerequisites are deliberately light. We assume you can read and write Python, and that you have met basic calculus, linear algebra, and probability: you should be comfortable with a derivative, a matrix-vector product \(A\mathbf{x}\), and the idea of a random variable with an expected value \(\mathbb{E}[X]\).
Everything else is built inside the book. Signals, sampling and the Nyquist condition, estimation, and a self-contained deep-learning refresher are all developed from scratch in Part I, with supporting derivations in the appendices. No prior signal-processing course and no prior machine-learning course is strictly required. If you have taken either, treat the relevant [F] sections as a fast review and move on.
Concretely, a reader arrives able to run this much, and the book supplies the rest:
import numpy as np
# You bring: NumPy, indexing, a mental model of a sampled signal.
fs = 1000 # sampling rate in Hz
t = np.arange(0, 1, 1 / fs) # one second of samples
x = np.sin(2 * np.pi * 5 * t) # a 5 Hz tone we later filter and classifyThe floor we assume: array creation and a sampled sinusoid. Filtering, spectral analysis, and learning on this signal are developed in Part I.
Read in order if the field is new to you; jump by tag if it is not. Either way, one book should carry you from a first measurement to a working perceptual system.