def img_to_frond_phase(folder,
label_folder,
dark=0,
avg=3,
dT=60,
p_range=12,
f_range=5,
save=""):
# 上の全部まとめたった！！
data = im.read_imgs(folder)
data = data.astype(float) - dark
label_img = im.read_imgs(label_folder)
# 解析をする．
imgs_phase = make_frond_phase_imgs(
imgs=data,
label_imgs=label_img,
avg=avg,
dT=dT,
p_range=p_range,
f_range=f_range,
save=save,
)
print(imgs_phase[0, 254, 254])
imgs_phase[(label_img != 0) * (np.isnan(imgs_phase))] = -2
print(imgs_phase[0, 254, 254])
color_phase = make_colors(imgs_phase, grey=-2, black=-1)
print(color_phase.shape)
if save is not False:
# color 画像の保存
im.save_imgs(save + "frond_phase_color", color_phase, extension="png")
return color_phase, imgs_phase
def frond_transform(
parent_directory,
calc_folder="mask_frond.tif",
other_folder_list=["mask_frond_lum.tif", "frond_lum.tif"],
motionType=1,
align=True,
):
"""移動補正を一括化
Args:
parent_directory: [description]
calc_folder: [description] (default: {"mask_frond"})
other_folder_list: [description] (default: {["mask_frond_lum", 'frond']})
motionType: [description] (default: {1})
"""
calc_imgs = im.read_imgs(os.path.join(parent_directory, calc_folder))
move_img, warps, _roops, trun_pi = imgs_transformECC(calc_imgs,
motionType=motionType,
align=align)
im.save_imgs(parent_directory, move_img, file_name="moved_" + calc_folder)
for i in other_folder_list:
other_imgs = im.read_imgs(os.path.join(parent_directory, i))
other_imgs = imgs_transformECC_warp(other_imgs, warps, trun_pi)
im.save_imgs(parent_directory, other_imgs, file_name="moved_" + i)
np.save(os.path.join(parent_directory, "warps.npy"), warps)
with open(os.path.join(parent_directory, "trun_pi"), mode="w") as f:
f.write(str(trun_pi))
def img_circadian_analysis(
folder,
avg=3,
f_avg=5,
mesh=1,
calc_range=[24, 24 * 3],
dt=60,
offset=0,
save=True,
tau_range=[16, 30],
background=0,
start_idx=0,
fft_nlls=True,
):
mask_frond = im.read_imgs(os.path.join(folder, "moved_mask_frond.tif"))
mask_frond = img_for_bio(mask_frond)
mask_frond[:start_idx, :, :] = 0
im.save_imgs(folder, mask_frond, "moved_mask_frond_usable.tif")
lum_folder = os.path.join(folder, "moved_mask_frond_lum.tif")
mask_folder = os.path.join(folder, "moved_mask_frond_usable.tif")
args = {
"folder": lum_folder,
"mask_folder": mask_folder,
"mesh": mesh,
"dt": dt,
"save": save,
}
img_pixel_theta(p_range=7,
f_avg=f_avg,
f_range=5,
make_color=[22, 28],
pdf=True,
xlsx=True,
r2_cut=0,
min_tau=16,
max_tau=30,
offset=offset,
background=background,
**args)
if fft_nlls is True:
if not os.path.exists(
os.path.join(
folder,
"FFT_nlls",
str(calc_range[0]) + "h-" + str(calc_range[1]) +
"h_mesh-" + str(mesh) + "_avg-" + str(avg),
)):
img_fft_nlls(calc_range=calc_range,
avg=avg,
tau_range=tau_range,
**args)
plt.savefig(os.path.join(pdf_save, str(i) + ".pdf"))
plt.close()
return 0
if __name__ == "__main__":
# グラフ化
os.chdir("/hdd1/kenya/Labo/keisan/python/00data")
# os.chdir('/home/kenya/Dropbox/Labo/python/00data')
days = ["./170613-LD2LL-ito-MVX", "./170829-LL2LL-ito-MVX", "./170215-LL2LL-MVX"]
for day in days:
frond_folder = day + "/frond"
data_file = "/small_phase_mesh1_avg3.npy"
for i in sorted(glob.glob(frond_folder + "/*")):
print(i)
frond = im.read_imgs(os.path.join(i, "small_moved_mask_frond_lum", ""))
area = np.empty(frond.shape[0])
lum_sum = np.empty(frond.shape[0])
for j in range(frond.shape[0]):
area[j] = np.count_nonzero(frond[j])
lum_sum[j] = np.sum(frond[j])
# 解析データのフォルダ
data = np.load(i + data_file)
save_folder = (
day
+ "/result/small_R_avg3/"
+ (data_file.lstrip("/")).rstrip(".npy")
+ str("_")
+ i.split("/")[-1]
)
if os.path.exists(day + "/result/small_R_avg3") is False:
label = np.array(change_frame.values).astype(np.int8)
colum = np.array(change_frame.columns).astype(np.int8)
print(label)
for i, x in enumerate(colum): # 置き換える値がx．それぞれの置き換える値に対して処理
for j, y in enumerate(label[:, i]): # 置き換えられる値がy それぞれの画像に対して処理
if y != 0: # 頭悪いのでおまじない．
new_label[j, imgs[j] == y] = x
return new_label
if __name__ == "__main__":
# os.chdir(os.path.join('/Users', 'kenya', 'keisan', 'python', '00data'))
os.chdir(
os.path.join("/hdd1", "Users", "kenya", "Labo", "keisan", "python",
"00data"))
# days = ('170613-LD2LL-ito-MVX')
days = "./170829-LL2LL-ito-MVX"
# days = ("170215-LL2LL-MVX")
today = datetime.datetime.today().strftime("%Y-%m-%d")
# days = ("./170215-LL2LL-MVX")
data_folder = os.path.join(days, "edit_raw")
img_folder = os.path.join(data_folder, "label_imgs_tmp.tif") # 元ラベル画像
change_file = os.path.join(data_folder, "label_190922.csv") # 変更に用いるファイル．
# ここから処理
imgs = im.read_imgs(img_folder, color=False)
change = pd.read_csv(change_file, index_col=0, header=0)
label_img = label_change(imgs, change)
im.save_imgs(os.path.join(data_folder), label_img,
"label_img" + today + ".tif")
img_after[img == k] = j + label_number[-1] + 1
imgs_centroids[i, j + label_number[-1]] = centroids[k - 1]
imgs_stats_sort[i, j + label_number[-1]] = stats[k - 1]
# 画像の格納庫に格納
imgs_sort[i] = img_after
label_out[i, :] = label_out[i, :] * np.in1d(label_out[i, :], img_after)
# label_out[label_out == 0] = np.nan
return imgs_sort, imgs_centroids, imgs_stats_sort, label_out
if __name__ == "__main__":
os.chdir(
os.path.join("/hdd1", "Users", "kenya", "Labo", "keisan", "python",
"00data"))
# day = "./170613-LD2LL-ito-MVX"
day = "./170829-LL2LL-ito-MVX"
# day = "./170215-LL2LL-MVX"
# ラベリングが終わった画像を取り込む．エッジと背景が0・それ以外は0で無い値なら良い
# 白黒でもカラーでもOK
frond_folder = os.path.join(day, "edit_raw", "div_ito_edit.tif")
# frond_folder = day + '/edit_raw/label_img_ito'
imgs = im.read_imgs(frond_folder)
imgs_sort, imgs_centroids, imgs_stats_sort, label_out = frond_traker(
imgs, min_size=5)
im.save_imgs(os.path.join(day, "edit_raw"), imgs_sort,
"label_imgs_tmp.tif")
label_out_save = pd.DataFrame(label_out,
columns=np.arange(1, label_out.shape[1] + 1))
label_out_save = label_out_save.replace(0, np.nan)
label_out_save.to_csv(os.path.join(day, "edit_raw", "label_tmp.csv"))
def img_fft_nlls(
folder,
mask_folder=False,
avg=3,
mesh=1,
dt=60,
save=True,
calc_range=[24, 24 * 3],
tau_range=[16, 30],
):
"""ピクセルごとに二次関数フィッティングにより解析する.
位相・周期・Peak時刻，推定の精度 を出力する．
Args:
folder: 画像群が入ったフォルダを指定する．
mask_folder: 背景が0になっている画像群を指定．
mesh: 解析前にメッシュ化する範囲 (default: {1} しない)
dt: Minite (default: {60})
save: 保存先のフォルダ名．Trueなら自動で名付け． (default: {False})
Returns:
保存のみの関数．返り値は持たさない(0)．
"""
if mesh == 1:
imgs = im.read_imgs(folder)
if mask_folder is not False:
mask = im.read_imgs(mask_folder)
imgs[mask == 0] = 0
else:
imgs = im.mesh_imgs(folder, mesh)
if mask_folder is not False:
mask = im.mesh_imgs(mask_folder, mesh)
imgs = imgs * (mask != 0).astype(np.float64)
##################################
# 生物発光の画像群から使うデータをcsvからと同じにする
##################################
# データの存在する場所のインデックスをとってくる)
use_xy = np.where(
np.all(
imgs[calc_range[0] * 60 // dt:calc_range[1] * 60 // dt + 1] != 0.0,
axis=0)) # データの存在する場所のインデックスをとってくる．
data = imgs[:, use_xy[0], use_xy[1]]
if data.shape[1] == 0:
print("解析できるデータがありません")
return False
print(str(data.shape[1]) + "個のデータを解析します")
########################################################
# 解析
########################################################
data_det, data_det_ampnorm = FFT_nlls.data_norm(data, dt=dt)
cosfit_args = {
"s": calc_range[0],
"e": calc_range[1],
"dt": dt,
"tau_range": tau_range,
}
fft_nlls_det = FFT_nlls.cos_fit(data_det, **cosfit_args)
fft_nlls_ampnorm = FFT_nlls.cos_fit(data_det_ampnorm, **cosfit_args)
###################################
# 周期を画像に戻す．
###################################
tau_dat_img = np.full_like(imgs[0], np.nan, dtype=np.float64)
tau_ampnorm_img = np.full_like(imgs[0], np.nan, dtype=np.float64)
rae_dat_img = np.full_like(imgs[0], np.nan, dtype=np.float64)
rae_ampnorm_img = np.full_like(imgs[0], np.nan, dtype=np.float64)
# 代入
tau_dat_img[use_xy] = fft_nlls_det["tau"]
rae_dat_img[use_xy] = fft_nlls_det["rae"]
tau_ampnorm_img[use_xy] = fft_nlls_ampnorm["tau"]
rae_ampnorm_img[use_xy] = fft_nlls_ampnorm["rae"]
# カラー画像に
tau_dat_img_colour = ((tau_dat_img - tau_range[0]) /
(tau_range[1] - tau_range[0]) * 0.7)
tau_ampnorm_img_colour = ((tau_ampnorm_img - tau_range[0]) /
(tau_range[1] - tau_range[0]) * 0.7)
color_legend = np.arange(30) * 0.7 / 30
color_legend = im.make_color(
np.vstack([color_legend, color_legend, color_legend]))
tau_dat_img_colour = im.make_color(tau_dat_img_colour, grey=-1)
tau_ampnorm_img_colour = im.make_color(tau_ampnorm_img_colour, grey=-1)
####################################
# 保存
####################
if save is True:
save = os.path.join(
os.path.dirname(folder),
"FFT_nlls",
str(calc_range[0]) + "h-" + str(calc_range[1]) + "h_mesh-" +
str(mesh) + "_avg-" + str(avg),
)
elif save is not False:
save = (save + "fft_nlls-" + str(calc_range[0]) + "-" +
str(calc_range[1]) + "_mesh-" + str(mesh) + "_avg-" + str(avg))
if save is not False:
if not os.path.exists(save):
os.makedirs(save)
# np.save(save + 'npy', tau_dat_img)
np.save(os.path.join(save, "fft_tau.npy"), tau_dat_img)
np.save(os.path.join(save, "fft_rae.npy"), rae_dat_img)
np.save(os.path.join(save, "fft_ampnorm_tau.npy"), tau_ampnorm_img)
np.save(os.path.join(save, "fft_ampnorm_rae.npy"), rae_ampnorm_img)
Image.fromarray(tau_dat_img_colour).save(save + ".tif")
Image.fromarray(tau_ampnorm_img_colour).save(save + "_ampnorm.tif")
return tau_dat_img, tau_ampnorm_img, tau_dat_img_colour, tau_ampnorm_img_colour
def img_pixel_theta(
folder,
mask_folder=False,
avg=3,
mesh=1,
dt=60,
offset=0,
p_range=7,
f_avg=1,
f_range=5,
background=0,
save=False,
make_color=[22, 28],
pdf=False,
xlsx=False,
distance_center=True,
r2_cut=0.5,
min_tau=16,
max_tau=32,
):
"""ピクセルごとに二次関数フィッティングにより解析する.
位相・周期・Peak時刻，推定の精度 を出力する．
Args:
folder: 画像群が入ったフォルダを指定する．
mask_folder: 背景が0になっている画像群を指定．
mesh: 解析前にメッシュ化する範囲 (default: {1} しない)
dt: Minite (default: {60})
offset: hour (default: {0})
p_range: 前後それぞれp_rangeよりも値が高い点をピークとみなす (default: {12})
f_avg: fiting前の移動平均 (default: {1})
f_range: 前後それぞれf_rangeのデータを用いて推定をする (default: {5})
save: 保存先のフォルダ名．Trueなら自動で名付け． (default: {False})
make_color: 周期を色付けする範囲 (default: {[22, 28]})
pdf: PDFを保存するか否か (default: {False})
xlsx: エクセルを保存するか否か (default: {False})
distance_center: 作図の際どこからの距離を取るか (default: {Ture} center)
Returns:
保存のみの関数．返り値は持たさない(0)．
"""
if save is True:
save = os.path.split(folder)[0]
save = os.path.join(
save,
"_".join([
"tau_mesh-" + str(mesh),
"avg-" + str(avg),
"prange-" + str(p_range),
"frange-" + str(f_range),
]),
)
##################################
# 画像の読み込
##################################
if mesh == 1:
imgs = im.read_imgs(folder)
if mask_folder is not False:
mask = im.read_imgs(mask_folder)
imgs[mask == 0] = 0
else:
mask = False
else:
imgs = im.mesh_imgs(folder, mesh)
if mask_folder is not False:
mask = im.mesh_imgs(mask_folder, mesh)
imgs[mask == 0] = 0
else:
mask = False
imgs = imgs.astype(np.float64)
imgs[imgs != 0] = imgs[imgs != 0] - background
##################################
# 生物発光の画像群から位相を求める
##################################
peak_a = make_theta_imgs(
imgs,
mask_img=mask,
avg=avg,
dt=dt,
p_range=p_range,
f_avg=f_avg,
f_range=f_range,
offset=offset,
r2_cut=r2_cut,
min_tau=min_tau,
max_tau=max_tau,
)
# データは見にくいので，カラーのデータを返す．
cv, sd, rms, avg_lum = peak_a[7], peak_a[8], peak_a[9], peak_a[10]
color_theta = im.make_colors(peak_a[0], grey=-1, black=np.nan)
###################################
# 保存用に周期を整形
###################################
tau_frond = peak_a[1] == -1
imgs_tau = (peak_a[1] - make_color[0]) / (make_color[1] -
make_color[0]) * 0.7
nan = ~np.isnan(imgs_tau)
imgs_tau[nan][imgs_tau[nan] > 0.7] = 0.7
imgs_tau[nan][imgs_tau[nan] < 0] = 0
imgs_tau[tau_frond] = -1
color_legend = np.arange(30) * 0.7 / 30
color_legend = im.make_color(
np.vstack([color_legend, color_legend, color_legend]))
color_tau = im.make_colors(imgs_tau, grey=-1)
####################################
# 保存用にampを整形
####################
def make_amp_color(data):
color = data / np.nanmax(data) * 0.7
color[data <= 0] = -1
color = im.make_colors(color, grey=-1)
return color
color_cv = make_amp_color(cv)
color_sd = make_amp_color(sd)
color_rms = make_amp_color(rms)
color_avg = make_amp_color(avg_lum)
####################################
# Peakの画像の作成
####################################
fold = int(60 / dt) * 24
p_img = make_peak_img(peak_a[0],
mask_imgs=mask,
idx_t=[0,
int(peak_a[0].shape[0] / fold) * fold + 1])
p_img = peak_img_list(p_img, per_ber=10, m=0, fold=fold)
if save is not False:
# color 画像の保存
def save_fnc(img, name=""):
"""名前つけが厄介なやつの保存"""
name = (name + str(np.nanmax(img)) + "-" +
str(np.nanmin(img[img > 0])) + ".tif")
im.save_imgs(save, img, name)
im.save_imgs(save, color_theta, "theta.tif")
save_fnc(color_tau, "tau_")
save_fnc(color_cv, "cv_")
save_fnc(color_sd, "sd_")
save_fnc(color_rms, "rms_")
save_fnc(color_avg, "move_avg_")
Image.fromarray(color_legend).save(os.path.join(
save, "color_legend.png"),
compress_level=0)
Image.fromarray(p_img).save(os.path.join(save, "peak_img.png"),
compress_level=0)
# phaseを0−1で表したものをcsvファイルに
np.save(os.path.join(save, "theta.npy"), peak_a[0])
np.save(os.path.join(save, "tau.npy"), imgs_tau)
np.save(os.path.join(save, "cv.npy"), cv)
np.save(os.path.join(save, "sd.npy"), sd)
np.save(os.path.join(save, "rms.npy"), rms)
np.save(os.path.join(save, "move_avg.npy"), avg_lum)
if pdf is not False:
if pdf is True:
pdf = os.path.join(save, "pdf")
else:
pdf = os.path.join(save, pdf)
img_analysis_pdf(
save_folder=pdf,
tau=peak_a[1],
cv=cv,
sd=sd,
distance_center=distance_center,
dt=dt,
)
if xlsx is not False:
writer = pd.ExcelWriter(os.path.join(save, "peak_list.xlsx"))
peak_a[2].T.to_excel(writer,
sheet_name="r2",
index=False,
header=True) # 保存
peak_a[5].T.to_excel(writer,
sheet_name="peak_time",
index=False,
header=True)
writer.save()
return 0
