Tune Peak Detection Sensitivity

find_peaks has several parameters that control which peaks are reported. This guide shows how to adjust them.

Problem

The default settings detect too many peaks (noise spikes) or too few (real peaks missed).

Solution

Construct some example data with two Lorentzian peaks and noise:

using OpticalSpectroscopy
using CurveFitModels

x = collect(range(2000, 2150, length=600))
y = lorentzian([0.45, 2062.0, 24.0, 0.0], x) .+
    lorentzian([0.30, 2110.0, 18.0, 0.0], x) .+
    0.005 .* randn(length(x))

Filter by Prominence

Prominence measures how much a peak stands out from its surroundings. The min_prominence parameter is a fraction of the data range (0 to 1):

# Default: keep peaks with prominence > 5% of data range
peaks = find_peaks(x, y)

# Stricter: only prominent peaks
peaks = find_peaks(x, y, min_prominence=0.1)

# More sensitive: catch small peaks too
peaks = find_peaks(x, y, min_prominence=0.02)

Filter by Width

Remove narrow spikes or overly broad features:

# Minimum width of 5 cm⁻¹
peaks = find_peaks(x, y, min_width=5.0)

# Both minimum and maximum width
peaks = find_peaks(x, y, min_width=5.0, max_width=100.0)

Adjust the Detection Window

The window parameter controls how many neighboring points a peak must exceed to count as a local maximum:

# Default: each peak must be higher than its immediate neighbors
peaks = find_peaks(x, y, window=1)

# Wider window: requires peak to dominate over ±3 points
peaks = find_peaks(x, y, window=3)

Larger windows suppress noise but may merge closely spaced peaks.

Apply Baseline Correction

Fluorescence background can create false peaks. Correct the baseline before detection:

peaks = find_peaks(x, y, baseline=:arpls)

# With custom baseline parameters
peaks = find_peaks(x, y, baseline=:snip, baseline_kw=(iterations=50,))

Iterative Workflow

Use find_peaks interactively to tune parameters before fitting:

# Start loose and tighten
peaks = find_peaks(x, y, baseline=:arpls)
println("$(length(peaks)) peaks found")
println(peak_table(peaks))

# Too many? Increase prominence
peaks = find_peaks(x, y, baseline=:arpls, min_prominence=0.1)
println("$(length(peaks)) peaks found")