def load_from_stats(self):
'''loads stats data and converts to timeseries without saving'''
stats = pd.read_csv(self.stats_filename, parse_dates=['timestamp'])
u = stats['timestamp'].unique()
u = pd.to_datetime(u)
sample_volume = scpp.get_sample_volume(
self.settings.PostProcess.pix_size,
path_length=self.settings.PostProcess.path_length)
dias, bin_lims = scpp.get_size_bins()
vd_oil = np.zeros((len(u), len(dias)))
vd_gas = np.zeros_like(vd_oil)
vd_total = np.zeros_like(vd_oil)
d50_gas = np.zeros(len(u))
d50_oil = np.zeros_like(d50_gas)
d50_total = np.zeros_like(d50_gas)
self.cos = np.zeros_like(d50_total)
# @todo make this number of particles per image, and sum according to index later
nparticles_all = 0
nparticles_total = 0
nparticles_oil = 0
nparticles_gas = 0
for i, s in enumerate(tqdm(u)):
substats = stats[stats['timestamp'] == s]
nparticles_all += len(substats)
nims = scpp.count_images_in_stats(substats)
sv = sample_volume * nims
oil = scog.extract_oil(substats)
nparticles_oil += len(oil)
dias, vd_oil_ = scpp.vd_from_stats(oil, self.settings.PostProcess)
vd_oil_ /= sv
vd_oil[i, :] = vd_oil_
gas = scog.extract_gas(substats)
nparticles_gas += len(gas)
dias, vd_gas_ = scpp.vd_from_stats(gas, self.settings.PostProcess)
vd_gas_ /= sv
vd_gas[i, :] = vd_gas_
d50_gas[i] = scpp.d50_from_vd(vd_gas_, dias)
nparticles_total += len(oil) + len(gas)
vd_total_ = vd_oil_ + vd_gas_
d50_total[i] = scpp.d50_from_vd(vd_total_, dias)
vd_total[i, :] = vd_total_
self.cos[i] = scog.cos_check(dias, vd_total[i, :])
self.vd_total = vd_total
self.vd_gas = vd_gas
self.vd_oil = vd_oil
self.d50_total = d50_total
self.d50_oil = d50_oil
self.d50_gas = d50_gas
self.u = u.tz_localize('UTC')
self.dias = dias
def convert_to_pj_format(stats_csv_file, config_file):
'''converts stats files into a total, and gas-only time-series csvfile which can be read by the old matlab
SummaryPlot exe'''
settings = PySilcamSettings(config_file)
logger.info('Loading stats....')
stats = pd.read_csv(stats_csv_file)
base_name = stats_csv_file.replace('-STATS.csv', '-PJ.csv')
gas_name = base_name.replace('-PJ.csv', '-PJ-GAS.csv')
ogdatafile = DataLogger(base_name, ogdataheader())
ogdatafile_gas = DataLogger(gas_name, ogdataheader())
stats['timestamp'] = pd.to_datetime(stats['timestamp'])
u = stats['timestamp'].unique()
sample_volume = sc_pp.get_sample_volume(
settings.PostProcess.pix_size,
path_length=settings.PostProcess.path_length)
logger.info('Analysing time-series')
for s in tqdm(u):
substats = stats[stats['timestamp'] == s]
nims = sc_pp.count_images_in_stats(substats)
sv = sample_volume * nims
oil = extract_oil(substats)
dias, vd_oil = sc_pp.vd_from_stats(oil, settings.PostProcess)
vd_oil /= sv
gas = extract_gas(substats)
dias, vd_gas = sc_pp.vd_from_stats(gas, settings.PostProcess)
vd_gas /= sv
d50_gas = sc_pp.d50_from_vd(vd_gas, dias)
vd_total = vd_oil + vd_gas
d50_total = sc_pp.d50_from_vd(vd_total, dias)
data_total = cat_data_pj(s, vd_total, d50_total, len(oil) + len(gas))
ogdatafile.append_data(data_total)
data_gas = cat_data_pj(s, vd_gas, d50_gas, len(gas))
ogdatafile_gas.append_data(data_gas)
logger.info(' OK.')
logger.info('Deleting header!')
with open(base_name, 'r') as fin:
data = fin.read().splitlines(True)
with open(base_name, 'w') as fout:
fout.writelines(data[1:])
with open(gas_name, 'r') as fin:
data = fin.read().splitlines(True)
with open(gas_name, 'w') as fout:
fout.writelines(data[1:])
logger.info('Conversion complete.')
def plot_trimmed_stats(self):
start_time = pd.to_datetime(
self.ui.dateTimeStart.dateTime().toPyDateTime())
end_time = pd.to_datetime(
self.ui.dateTimeEnd.dateTime().toPyDateTime())
self.trimmed_stats, self.output_filename = scpp.trim_stats(
self.stats_filename,
start_time,
end_time,
write_new=False,
stats=self.stats)
if np.isnan(self.trimmed_stats.equivalent_diameter.max()) or len(
self.trimmed_stats) == 0:
QMessageBox.warning(
self, "No data in this segment!",
'No data was found within the specified time range.',
QMessageBox.Ok)
return
settings = PySilcamSettings(self.config_file)
plt.figure(self.figure.number)
plt.clf()
stats_oil = scog.extract_oil(self.trimmed_stats)
stats_gas = scog.extract_gas(self.trimmed_stats)
sample_volume = scpp.get_sample_volume(
settings.PostProcess.pix_size,
path_length=settings.PostProcess.path_length)
nims = scpp.count_images_in_stats(self.trimmed_stats)
dias, vd = scpp.vd_from_stats(self.trimmed_stats, settings.PostProcess)
dias, vd_oil = scpp.vd_from_stats(stats_oil, settings.PostProcess)
dias, vd_gas = scpp.vd_from_stats(stats_gas, settings.PostProcess)
sv = sample_volume * nims
vd /= sv
vd_oil /= sv
vd_gas /= sv
plt.plot(dias, vd_oil + vd_gas, 'k', label='TOTAL')
plt.plot(dias, vd_oil, 'r', label='OIL')
plt.plot(dias, vd_gas, 'b', label='GAS')
plt.xscale('log')
plt.xlabel('Equiv. diam (um)')
plt.ylabel('Volume concentration (uL/L)')
plt.xlim(10, 12000)
plt.legend()
self.canvas.draw()
def test_analysis(img, PIX_SIZE, PATH_LENGTH, config_file=''):
'''wrapper for pysilcam processing
Args:
img (unit8) : image to be processed (equivalent to the background-corrected image obtained from the SilCam)
PIX_SIZE (float) : pixel size of the setup [um]
PATH_LENGTH (float) : path length of the setup [mm]
Returns:
dias (array) : mid points of the size distribution bins [um]
vd (array) : volume concentration in each size bin [uL/L/bin]
imbw (uint8) : segmented image
stat_extract_time (timestamp) : time taken to run statextract
'''
# administer configuration settings according to specified setup
if config_file=='':
testconfig = os.path.split(sccf.default_config_path())[0]
testconfig = os.path.join(testconfig, 'tests/config.ini')
conf.set('Process', 'real_time_stats', 'True')
conf.set('Process', 'threshold', '0.85')
conf.set('ExportParticles', 'export_images', 'False')
else:
testconfig = config_file
conf = sccf.load_config(testconfig)
conf.set('PostProcess', 'pix_size', str(PIX_SIZE))
conf.set('PostProcess', 'path_length', str(PATH_LENGTH))
settings = sccf.PySilcamSettings(conf) # pass these settings without saving a config file to disc
# load tensorflow model
nnmodel, class_labels = sccl.load_model(model_path=settings.NNClassify.model_path)
start_time = pd.Timestamp.now() # time statextract
# process the image
stats, imbw, saturation = scpr.statextract(img, settings, pd.Timestamp.now(),
nnmodel, class_labels)
end_time = pd.Timestamp.now() # time statextract
stat_extract_time = end_time - start_time
dias, vd = scpp.vd_from_stats(stats, settings.PostProcess) # calculate the volume distribution from the processed stats
# scale the volume distribution according to the SilCam setup specified
sv = scpp.get_sample_volume(settings.PostProcess.pix_size,
path_length=settings.PostProcess.path_length)
vd /= sv
return dias, vd, imbw, stat_extract_time
def nd_scaled(stats, settings, ax, c='k'):
'''
Plot the particle number distribution, scaled to the total volume of water sampled
Args:
stats (DataFrame) : particle statistics from silcam process
settings (PySilcamSettings) : settings associated with the data, loaded with PySilcamSettings
ax () : axis to plot data on
c='k' (str) : color of the line to be plotted
'''
sv = sc_pp.get_sample_volume(settings.pix_size,
path_length=settings.path_length,
imx=2048, imy=2448) # sample volume per image
# re-scale sample volume to number of images
sv_total = sv * sc_pp.count_images_in_stats(stats) # total sampled water volume
nd(stats, settings, ax, line=None, c='k', sample_volume=sv_total)
return
def gor_timeseries(stats, settings):
from tqdm import tqdm
u = stats['timestamp'].unique()
td = pd.to_timedelta('00:00:' + str(settings.window_size / 2.))
sample_volume = sc_pp.get_sample_volume(settings.pix_size,
path_length=settings.path_length)
gor = []
time = []
for t in tqdm(u):
dt = pd.to_datetime(t)
stats_ = stats[(pd.to_datetime(stats['timestamp']) < (dt + td))
& (pd.to_datetime(stats['timestamp']) > (dt - td))]
oilstats = extract_oil(stats_)
dias, vd_oil = sc_pp.vd_from_stats(oilstats, settings)
nims = sc_pp.count_images_in_stats(oilstats)
sv = sample_volume * nims
vd_oil /= sv
gasstats = extract_gas(stats_)
dias, vd_gas = sc_pp.vd_from_stats(gasstats, settings)
nims = sc_pp.count_images_in_stats(gasstats)
sv = sample_volume * nims
vd_gas /= sv
gor_ = sum(vd_gas) / sum(vd_oil)
time.append(pd.to_datetime(t))
gor.append(gor_)
if (len(gor) == 0) or (np.isnan(max(gor))):
gor = np.nan
time = np.nan
return gor, time
def export_timeseries(configfile, statsfile):
settings = PySilcamSettings(configfile)
print('Loading STATS data: ', statsfile)
stats = pd.read_csv(statsfile)
stats['timestamp'] = pd.to_datetime(stats['timestamp'])
stats.sort_values(by='timestamp', inplace=True)
print('Extracting oil and gas')
stats_oil = scog.extract_oil(stats)
stats_gas = scog.extract_gas(stats)
print('Calculating timeseries')
u = pd.to_datetime(stats['timestamp']).unique()
sample_volume = sc_pp.get_sample_volume(settings.PostProcess.pix_size, path_length=settings.PostProcess.path_length)
td = pd.to_timedelta('00:00:' + str(settings.PostProcess.window_size / 2.))
vdts_all = []
vdts_oil = []
vdts_gas = []
d50_all = []
d50_oil = []
d50_gas = []
timestamp = []
d50_av_all = []
d50_av_oil = []
d50_av_gas = []
gor = []
for s in tqdm(u):
timestamp.append(pd.to_datetime(s))
dt = pd.to_datetime(s)
dias, vd_all = sc_pp.vd_from_stats(stats[stats['timestamp'] == s],
settings.PostProcess)
dias, vd_oil = sc_pp.vd_from_stats(stats_oil[stats_oil['timestamp'] == s],
settings.PostProcess)
dias, vd_gas = sc_pp.vd_from_stats(stats_gas[stats_gas['timestamp'] == s],
settings.PostProcess)
nims = sc_pp.count_images_in_stats(stats[stats['timestamp'] == s])
sv = sample_volume * nims
vd_all /= sv
vd_oil /= sv
vd_gas /= sv
d50_all.append(sc_pp.d50_from_vd(vd_all, dias))
d50_oil.append(sc_pp.d50_from_vd(vd_oil, dias))
d50_gas.append(sc_pp.d50_from_vd(vd_gas, dias))
vdts_all.append(vd_all)
vdts_oil.append(vd_oil)
vdts_gas.append(vd_gas)
stats_av = stats[(stats['timestamp']<(dt+td)) & (stats['timestamp']>(dt-td))]
stats_av_oil = scog.extract_oil(stats_av)
stats_av_gas = scog.extract_gas(stats_av)
d50_av_all.append(sc_pp.d50_from_stats(stats_av, settings.PostProcess))
d50_av_oil.append(sc_pp.d50_from_stats(stats_av_oil, settings.PostProcess))
d50_av_gas.append(sc_pp.d50_from_stats(stats_av_gas, settings.PostProcess))
dias, vdts_av = sc_pp.vd_from_stats(stats_av, settings.PostProcess)
dias, vdts_av_oil = sc_pp.vd_from_stats(stats_av_oil, settings.PostProcess)
dias, vdts_av_gas = sc_pp.vd_from_stats(stats_av_gas, settings.PostProcess)
nims = sc_pp.count_images_in_stats(stats_av)
sv = sample_volume * nims
vdts_av /= sv
vdts_av_oil /= sv
vdts_av_gas /= sv
gor.append(np.sum(vdts_av_gas)/np.sum(vdts_av_oil))
outpath, outfile = os.path.split(statsfile)
outfile = outfile.replace('-STATS.csv','')
outfile = os.path.join(outpath, outfile)
time_series = pd.DataFrame(data=np.squeeze(vdts_all), columns=dias)
time_series['D50'] = d50_all
time_series['Time'] = timestamp
time_series.to_excel(outfile +
'-TIMESERIES' + '' + '.xlsx')
time_series = pd.DataFrame(data=np.squeeze(vdts_oil), columns=dias)
time_series['D50'] = d50_oil
time_series['Time'] = timestamp
time_series.to_excel(outfile +
'-TIMESERIES' + 'oil' + '.xlsx')
time_series = pd.DataFrame(data=np.squeeze(vdts_gas), columns=dias)
time_series['D50'] = d50_gas
time_series['Time'] = timestamp
time_series.to_excel(outfile +
'-TIMESERIES' + 'gas' + '.xlsx')
plt.figure(figsize=(20, 10))
if not np.min(np.isnan(d50_oil)):
plt.plot(timestamp, d50_oil, 'ro')
if not np.min(np.isnan(d50_av_oil)):
plt.plot(timestamp, d50_av_oil, 'r-')
lns1 = plt.plot(np.nan, np.nan, 'r-', label='OIL')
if not np.min(np.isnan(d50_gas)):
plt.plot(timestamp, d50_gas, 'bo')
if not np.min(np.isnan(d50_av_gas)):
plt.plot(timestamp, d50_av_gas, 'b-')
lns2 = plt.plot(np.nan, np.nan, 'b-', label='GAS')
plt.ylabel('d50 [um]')
plt.ylim(0, max(plt.gca().get_ylim()))
ax = plt.gca().twinx()
plt.sca(ax)
plt.ylabel('GOR')
if not np.min(np.isnan(gor)):
plt.plot(timestamp, gor, 'k')
lns3 = plt.plot(np.nan, np.nan, 'k', label='GOR')
plt.ylim(0, max(plt.gca().get_ylim()))
lns = lns1 + lns2 + lns3
labs = [l.get_label() for l in lns]
plt.legend(lns, labs)
plt.savefig(outfile +
'-d50_TimeSeries.png', dpi=600, bbox_inches='tight')
plt.close()
print('Export figure made. ')
print('Exporting averages... ')
# average all
dias, vd = sc_pp.vd_from_stats(stats,
settings.PostProcess)
nims = sc_pp.count_images_in_stats(stats)
sv = sample_volume * nims
vd /= sv
d50 = sc_pp.d50_from_vd(vd, dias)
dfa = pd.DataFrame(data=[vd], columns=dias)
dfa['d50'] = d50
timestamp = np.min(pd.to_datetime(stats['timestamp']))
dfa['Time'] = timestamp
dfa.to_excel(statsfile.replace('-STATS.csv', '') +
'-AVERAGE' + '' + '.xlsx')
#average oil
dias, vd = sc_pp.vd_from_stats(stats_oil,
settings.PostProcess)
vd /= sv # sample volume remains the same as 'all'
d50 = sc_pp.d50_from_vd(vd, dias)
dfa = pd.DataFrame(data=[vd], columns=dias)
dfa['d50'] = d50
timestamp = np.min(pd.to_datetime(stats['timestamp'])) # still use total stats for this time
dfa['Time'] = timestamp
dfa.to_excel(statsfile.replace('-STATS.csv', '') +
'-AVERAGE' + 'oil' + '.xlsx')
#average gas
dias, vd = sc_pp.vd_from_stats(stats_gas,
settings.PostProcess)
vd /= sv # sample volume remains the same as 'all'
d50 = sc_pp.d50_from_vd(vd, dias)
dfa = pd.DataFrame(data=[vd], columns=dias)
dfa['d50'] = d50
timestamp = np.min(pd.to_datetime(stats['timestamp'])) # still use total stats for this time
dfa['Time'] = timestamp
dfa.to_excel(statsfile.replace('-STATS.csv', '') +
'-AVERAGE' + 'gas' + '.xlsx')
print('Export done: ', outfile)
def generate_report(report_name, PIX_SIZE = 28.758169934640524,
PATH_LENGTH=40, d50 = 400, TotalVolumeConcentration = 800,
MinD = 108, config_file=''):
'''Create a report of the expected response of the silcam to the provided experimental setup
Args:
report_name (str) : The path and filename of a pdf to be created
PIX_SIZE (float) : pixel size of the setup [um]
PATH_LENGTH (float) : the path length of the setup [mm] Path length is the gap between housings
d50 (float) : the expected of the oil d50 (50th percentile of the cumulative sum of the volume distribution)
TotalVolumeConcentration (float) : the expected concentration of oil in the sample volume [uL/L]
MinD (float) : minimum resolvable diameter of the setup [um]. this would usually scale with the pixel size.
synthesized particles smaller than this are also removed for speed purposes
'''
plt.close('all')
pp = PdfPages(report_name)
# image dimensions (fixed always for GC2450 camera)
imx = 2448
imy = 2048
# get diameters and limits of size bins
diams, bin_limits_um = scpp.get_size_bins()
# initial volume distribution, close to Oystein's MPB paper in 2013
vd = weibull(diams, n=d50)
vd = vd/np.sum(vd)*TotalVolumeConcentration # scale the distribution according to concentration
DropletVolume=((4/3)*np.pi*((diams*1e-6)/2)**3) # the volume of each droplet in m3
nd=vd/(DropletVolume*1e9) # the number distribution in each bin
nd[diams<MinD] = 0 # remove small particles for speed purposes
# calculate the sample volume of the SilCam specified
sv = scpp.get_sample_volume(PIX_SIZE, path_length=PATH_LENGTH, imx=imx, imy=imy)
nd = nd*sv # scale the number distribution by the sample volume so resulting units are #/L/bin
nc = int(sum(nd)) # calculate the total number concentration
vd2 = scpp.vd_from_nd(nd,diams,sv) # convert the number distribution to volume distribution in uL/L/bin
vc_initial = sum(vd2) # obtain the resulting concentration, now having remove small particles
d50_theory = scpp.d50_from_vd(vd2, diams) # calculate the d50 in um
plt.plot(diams, vd2, 'k', label='Initial')
plt.plot(diams, vd, 'r:', label='Theoretical')
plt.vlines(d50_theory, 0, max(vd2), linestyle='--')
plt.xscale('log')
plt.xlabel('ECD [um]')
plt.xlabel('Volume distribution [uL/L/bin]')
plt.legend()
plt.title('Initial conditions:' +
'\n\n' + str(nc) + ' particles per image volume' +
'\n' + str(int(vc_initial)) + ' initial volume concentration [uL/L]' +
'\n' + str(int(d50_theory)) + ' initial d50 [um]',
horizontalalignment='left', loc='left')
pp.savefig(bbox_inches='tight')
nims = 40 # the number of images to simulate
# preallocate variables
log_vd = np.zeros((nims,len(diams)))
cvd = np.zeros(nims)
cd50 = np.zeros(nims)
for I in range(nims):
# randomly select a droplet radius from the input distribution
rad = np.random.choice(diams/2, size=nc, p=nd/sum(nd)) / PIX_SIZE # radius is in pixels
log_ecd = rad*2*PIX_SIZE # log this size as a diameter in um
necd, edges = np.histogram(log_ecd,bin_limits_um) # count particles into number distribution
log_vd[I,:] = scpp.vd_from_nd(necd,diams) # convert to volume distribution
cvd[I] = np.sum(np.mean(log_vd[0:I,:],axis=0)) # calculated the cumulate volume distribution over image number
cd50[I] = scpp.d50_from_vd(np.mean(log_vd,axis=0), diams) # calcualte the cumulate d50 over image number
f, a = plt.subplots(1,3,figsize=(16,4))
plt.sca(a[0])
plt.plot(diams, vd2, 'k')
plt.plot(diams, log_vd[0,:]/sv, alpha=0.5, label='1 image')
plt.plot(diams, np.mean(log_vd[0:4,:], axis=0)/sv, alpha=0.5, label='4 images')
plt.plot(diams, np.mean(log_vd, axis=0)/sv, alpha=0.5, label=(str(nims) + ' images'))
plt.xscale('log')
plt.vlines(d50_theory, 0, max(vd2), linestyle='--')
plt.xlabel('ECD [um]')
plt.ylabel('Volume distribution [uL/L/bin]')
plt.legend()
plt.title('Statistical summaries')
plt.sca(a[1])
plt.plot(cvd/sv,'k')
plt.hlines(vc_initial, 0, nims, linestyle='--')
plt.xlabel('Image number')
plt.ylabel('Volume concentration [uL/L]')
plt.sca(a[2])
plt.plot(cd50,'k')
plt.hlines(d50_theory, 0, nims, linestyle='--')
plt.xlabel('Image number')
plt.ylabel('d50 [um]')
pp.savefig(bbox_inches='tight')
# synthesize an image, returning the segmented image and the inputted volume distribution
img, log_vd = synthesize(diams, bin_limits_um, nd, imx, imy, PIX_SIZE)
plt.figure(figsize=(10,10))
plt.imshow(img, vmin=0, vmax=255, extent=[0,imx*PIX_SIZE/1000,0,imy*PIX_SIZE/1000])
plt.xlabel('[mm]')
plt.ylabel('[mm]')
plt.title('Synthetic image')
pp.savefig(bbox_inches='tight')
vd = np.zeros_like(log_vd)
imbw = np.zeros_like(img[:,:,0])
stat_extract_time = pd.Timedelta(seconds=0)
# @todo this should be handles properly as part of testing
try:
diams, vd, imbw, stat_extract_time = test_analysis(img, PIX_SIZE, PATH_LENGTH, config_file=config_file)
except:
print('Analysis failed')
pass
f, a = plt.subplots(1,2,figsize=(20,8))
plt.sca(a[0])
plt.plot(diams, vd2, 'r:', label='Initial')
plt.plot(diams, log_vd/sv ,'k', label='Statistical Best')
plt.plot(diams, vd, 'g', alpha=0.5, label='PySilCam')
plt.xscale('log')
plt.xlabel('ECD [um]')
plt.ylabel('Volume distribution [uL/L]')
plt.legend()
plt.title('Single image assessment:' +
'\n\n' + 'Statextract took ' + str(stat_extract_time.seconds) + ' seconds',
horizontalalignment='left', loc='left')
plt.sca(a[1])
plt.imshow(imbw, vmin=0, vmax=1, extent=[0,imx*PIX_SIZE/1000,0,imy*PIX_SIZE/1000], cmap='gray')
plt.xlabel('[mm]')
plt.ylabel('[mm]')
plt.title('imbw')
pp.savefig(bbox_inches='tight')
pp.close()
