Changes

              
    

          
          
        
        

            
f
1
{
f
1
{
            
2
  "author": null,
2
  "author": null,
            
3
  "author_email": null,
3
  "author_email": null,
            
4
  "code": "7b.b.0c",
4
  "code": "7b.b.0c",
            
5
  "contributor": "NIOZ Royal Netherlands Institute for Sea Research",
5
  "contributor": "NIOZ Royal Netherlands Institute for Sea Research",
            
6
  "creator_user_id": "26b029ba-1ef0-4577-a666-580bc1a26c0f",
6
  "creator_user_id": "26b029ba-1ef0-4577-a666-580bc1a26c0f",
            
7
  "creators": 
7
  "creators": 
            
8
tname\":\"Jeroen\",\"lastname\":\"Hoekendijk\",\"affiliation\":\"Royal 
8
tname\":\"Jeroen\",\"lastname\":\"Hoekendijk\",\"affiliation\":\"Royal 
            
9
Netherlands Institute for Sea 
9
Netherlands Institute for Sea 
            
10
ess\":\"Jeroen.hoekendijk@nioz.nl\",\"iscorrespondingauthor\":true}]",
10
ess\":\"Jeroen.hoekendijk@nioz.nl\",\"iscorrespondingauthor\":true}]",
            
11
  "dataset_persistent_id": "DOI:10.33591/nioz/7b.b.0c",
11
  "dataset_persistent_id": "DOI:10.33591/nioz/7b.b.0c",
            
12
  "dates": "",
12
  "dates": "",
            
13
  "deposit_date": "2021-11-01",
13
  "deposit_date": "2021-11-01",
            
14
  "depositor": "Jeroen Hoekendijk",
14
  "depositor": "Jeroen Hoekendijk",
            
15
  "distribution_date": "2021-11-01",
15
  "distribution_date": "2021-11-01",
            
16
  "distributor": "Research Data Management(NIOZ Royal Netherlands 
16
  "distributor": "Research Data Management(NIOZ Royal Netherlands 
            
17
Institute for Sea Research)",
17
Institute for Sea Research)",
            
18
  "doi_date_published": "2021-12-01",
18
  "doi_date_published": "2021-12-01",
            
19
  "funding_references": "[{\"name\":\"NWO \",\"awardnumber\":\"project 
19
  "funding_references": "[{\"name\":\"NWO \",\"awardnumber\":\"project 
            
20
ALWPP.2017.003\",\"awardtitle\":\"\"}]",
20
ALWPP.2017.003\",\"awardtitle\":\"\"}]",
            
21
  "groups": [],
21
  "groups": [],
            
22
  "id": "d3383e5d-fd05-4800-9df3-d914770fbf0e",
22
  "id": "d3383e5d-fd05-4800-9df3-d914770fbf0e",
            
23
  "isopen": false,
23
  "isopen": false,
            
24
  "license_title": null,
24
  "license_title": null,
            
25
  "maintainer": null,
25
  "maintainer": null,
            
26
  "maintainer_email": null,
26
  "maintainer_email": null,
            
27
  "metadata_created": "2024-10-22T11:45:18.547538",
27
  "metadata_created": "2024-10-22T11:45:18.547538",
            
n
28
  "metadata_modified": "2024-10-22T11:45:18.834524",
n
28
  "metadata_modified": "2024-10-22T11:45:19.264357",
            
29
  "name": "7bb0c",
29
  "name": "7bb0c",
            
30
  "notes": "Many ecological studies rely on count data and involve 
30
  "notes": "Many ecological studies rely on count data and involve 
            
31
manual counting of objects of interest, which is time-consuming and 
31
manual counting of objects of interest, which is time-consuming and 
            
32
especially disadvantageous when time in the field or lab is limited. 
32
especially disadvantageous when time in the field or lab is limited. 
            
33
However, an increasing number of works uses digital imagery, which 
33
However, an increasing number of works uses digital imagery, which 
            
34
opens opportunities to automatise counting tasks. In this study, we 
34
opens opportunities to automatise counting tasks. In this study, we 
            
35
use machine learning to automate counting objects of interest without 
35
use machine learning to automate counting objects of interest without 
            
36
the need to label individual objects. By leveraging already existing 
36
the need to label individual objects. By leveraging already existing 
            
37
image-level annotations, this approach can also give value to 
37
image-level annotations, this approach can also give value to 
            
38
historical data that were collected and annotated over longer time 
38
historical data that were collected and annotated over longer time 
            
39
series (typical for many ecological studies), without the aim of deep 
39
series (typical for many ecological studies), without the aim of deep 
            
40
learning applications. We demonstrate deep learning regression on two 
40
learning applications. We demonstrate deep learning regression on two 
            
41
fundamentally different counting tasks: (i) daily growth rings from 
41
fundamentally different counting tasks: (i) daily growth rings from 
            
42
microscopic images of fish otolith (i.e., hearing stone) and (ii) 
42
microscopic images of fish otolith (i.e., hearing stone) and (ii) 
            
43
hauled out seals from highly variable aerial imagery. In the otolith 
43
hauled out seals from highly variable aerial imagery. In the otolith 
            
44
images, our deep learning-based regressor yields an RMSE of 3.40 
44
images, our deep learning-based regressor yields an RMSE of 3.40 
            
45
day-rings and an R^2 of 0.92. Initial performance in the seal images 
45
day-rings and an R^2 of 0.92. Initial performance in the seal images 
            
46
is lower (RMSE of 23.46 seals and R^2 of 0.72), which can be 
46
is lower (RMSE of 23.46 seals and R^2 of 0.72), which can be 
            
47
attributed to a lack of images with a high number of seals in the 
47
attributed to a lack of images with a high number of seals in the 
            
48
initial training set, compared to the test set. We then show how to 
48
initial training set, compared to the test set. We then show how to 
            
49
improve performance substantially (RMSE of 19.03 seals and R^2 of 
49
improve performance substantially (RMSE of 19.03 seals and R^2 of 
            
50
0.77) by carefully selecting and relabelling just 100 additional 
50
0.77) by carefully selecting and relabelling just 100 additional 
            
51
training images based on initial model prediction discrepancy. The 
51
training images based on initial model prediction discrepancy. The 
            
52
regression-based approach used here returns accurate counts (R^2 of 
52
regression-based approach used here returns accurate counts (R^2 of 
            
53
0.92 and 0.77 for the rings and seals, respectively), directly usable 
53
0.92 and 0.77 for the rings and seals, respectively), directly usable 
            
54
in ecological research.",
54
in ecological research.",
            
n
55
  "num_resources": 1,
n
55
  "num_resources": 2,
            
56
  "num_tags": 9,
56
  "num_tags": 9,
            
57
  "organization": {
57
  "organization": {
            
58
    "approval_status": "approved",
58
    "approval_status": "approved",
            
59
    "created": "2024-10-22T06:06:56.074257",
59
    "created": "2024-10-22T06:06:56.074257",
            
60
    "description": "",
60
    "description": "",
            
61
    "id": "bed2a893-3895-450b-bf4b-c74318003b1b",
61
    "id": "bed2a893-3895-450b-bf4b-c74318003b1b",
            
62
    "image_url": 
62
    "image_url": 
            
63
z.nl/application/files/1216/8924/5333/NIOZ-logo_witte_randi-2023.svg",
63
z.nl/application/files/1216/8924/5333/NIOZ-logo_witte_randi-2023.svg",
            
64
    "is_organization": true,
64
    "is_organization": true,
            
65
    "name": "nioz",
65
    "name": "nioz",
            
66
    "state": "active",
66
    "state": "active",
            
67
    "title": "Royal Netherlands Institute for Sea Research",
67
    "title": "Royal Netherlands Institute for Sea Research",
            
68
    "type": "organization"
68
    "type": "organization"
            
69
  },
69
  },
            
70
  "owner_org": "bed2a893-3895-450b-bf4b-c74318003b1b",
70
  "owner_org": "bed2a893-3895-450b-bf4b-c74318003b1b",
            
71
  "private": false,
71
  "private": false,
            
72
  "production_date": "2021-12-01",
72
  "production_date": "2021-12-01",
            
73
  "relationships_as_object": [],
73
  "relationships_as_object": [],
            
74
  "relationships_as_subject": [],
74
  "relationships_as_subject": [],
            
75
  "resources": [
75
  "resources": [
            
76
    {
76
    {
            
77
      "cache_last_updated": null,
77
      "cache_last_updated": null,
            
78
      "cache_url": null,
78
      "cache_url": null,
            
79
      "created": "2024-10-22T11:45:18.861403",
79
      "created": "2024-10-22T11:45:18.861403",
            
80
      "datastore_active": false,
80
      "datastore_active": false,
            
81
      "format": "CSV",
81
      "format": "CSV",
            
82
      "hash": "b05d852ac7aa1a15c1c2d458bb331d7c",
82
      "hash": "b05d852ac7aa1a15c1c2d458bb331d7c",
            
83
      "id": "3f0c9e4a-9993-4082-9a84-118d65611ec9",
83
      "id": "3f0c9e4a-9993-4082-9a84-118d65611ec9",
            
84
      "last_modified": "2024-10-22T11:45:18.825836",
84
      "last_modified": "2024-10-22T11:45:18.825836",
            
n
85
      "metadata_modified": "2024-10-22T11:45:18.841101",
n
85
      "metadata_modified": "2024-10-22T11:45:19.269937",
            
86
      "mimetype": "text/csv",
86
      "mimetype": "text/csv",
            
87
      "mimetype_inner": null,
87
      "mimetype_inner": null,
            
88
      "name": "Labels_Seal_subset_1.csv",
88
      "name": "Labels_Seal_subset_1.csv",
            
89
      "package_id": "d3383e5d-fd05-4800-9df3-d914770fbf0e",
89
      "package_id": "d3383e5d-fd05-4800-9df3-d914770fbf0e",
            
90
      "position": 0,
90
      "position": 0,
            
91
      "resource_type": null,
91
      "resource_type": null,
            
92
      "size": 14487,
92
      "size": 14487,
            
93
      "state": "active",
93
      "state": "active",
            
94
      "url": 
94
      "url": 
            
95
0c9e4a-9993-4082-9a84-118d65611ec9/download/labels_seal_subset_1.csv",
95
0c9e4a-9993-4082-9a84-118d65611ec9/download/labels_seal_subset_1.csv",
            
t
t
96
      "url_type": "upload"
            
97
    },
            
98
    {
            
99
      "cache_last_updated": null,
            
100
      "cache_url": null,
            
101
      "created": "2024-10-22T11:45:19.291807",
            
102
      "datastore_active": false,
            
103
      "format": "TXT",
            
104
      "hash": "d9e3323d4134508a4c6eddc590882e95",
            
105
      "id": "c0027676-6c2a-41f5-8e9b-97c17d3eeb13",
            
106
      "last_modified": "2024-10-22T11:45:19.239911",
            
107
      "metadata_modified": "2024-10-22T11:45:19.271029",
            
108
      "mimetype": "text/plain",
            
109
      "mimetype_inner": null,
            
110
      "name": "README.txt",
            
111
      "package_id": "d3383e5d-fd05-4800-9df3-d914770fbf0e",
            
112
      "position": 1,
            
113
      "resource_type": null,
            
114
      "size": 2343,
            
115
      "state": "active",
            
116
      "url": 
            
117
0e/resource/c0027676-6c2a-41f5-8e9b-97c17d3eeb13/download/readme.txt",
            
96
      "url_type": "upload"
118
      "url_type": "upload"
            
97
    }
119
    }
            
98
  ],
120
  ],
            
99
  "state": "active",
121
  "state": "active",
            
100
  "subject": "Earth and Environmental Sciences",
122
  "subject": "Earth and Environmental Sciences",
            
101
  "tags": [
123
  "tags": [
            
102
    {
124
    {
            
103
      "display_name": "Aerial Surveys",
125
      "display_name": "Aerial Surveys",
            
104
      "id": "d1997720-7e21-4688-b8d9-53931f0fa407",
126
      "id": "d1997720-7e21-4688-b8d9-53931f0fa407",
            
105
      "name": "Aerial Surveys",
127
      "name": "Aerial Surveys",
            
106
      "state": "active",
128
      "state": "active",
            
107
      "vocabulary_id": null
129
      "vocabulary_id": null
            
108
    },
130
    },
            
109
    {
131
    {
            
110
      "display_name": "Computer Vision",
132
      "display_name": "Computer Vision",
            
111
      "id": "6b8c23bb-a300-4c6e-8ac1-7b3ae58f3ecb",
133
      "id": "6b8c23bb-a300-4c6e-8ac1-7b3ae58f3ecb",
            
112
      "name": "Computer Vision",
134
      "name": "Computer Vision",
            
113
      "state": "active",
135
      "state": "active",
            
114
      "vocabulary_id": null
136
      "vocabulary_id": null
            
115
    },
137
    },
            
116
    {
138
    {
            
117
      "display_name": "Counting",
139
      "display_name": "Counting",
            
118
      "id": "5c5fdd29-e46c-47f8-8533-b9280d035653",
140
      "id": "5c5fdd29-e46c-47f8-8533-b9280d035653",
            
119
      "name": "Counting",
141
      "name": "Counting",
            
120
      "state": "active",
142
      "state": "active",
            
121
      "vocabulary_id": null
143
      "vocabulary_id": null
            
122
    },
144
    },
            
123
    {
145
    {
            
124
      "display_name": "Deep Learning",
146
      "display_name": "Deep Learning",
            
125
      "id": "39730ab2-462d-4f54-92b7-11d18dcae180",
147
      "id": "39730ab2-462d-4f54-92b7-11d18dcae180",
            
126
      "name": "Deep Learning",
148
      "name": "Deep Learning",
            
127
      "state": "active",
149
      "state": "active",
            
128
      "vocabulary_id": null
150
      "vocabulary_id": null
            
129
    },
151
    },
            
130
    {
152
    {
            
131
      "display_name": "Ecology",
153
      "display_name": "Ecology",
            
132
      "id": "ad553155-23ec-4177-befc-34a9bf75d351",
154
      "id": "ad553155-23ec-4177-befc-34a9bf75d351",
            
133
      "name": "Ecology",
155
      "name": "Ecology",
            
134
      "state": "active",
156
      "state": "active",
            
135
      "vocabulary_id": null
157
      "vocabulary_id": null
            
136
    },
158
    },
            
137
    {
159
    {
            
138
      "display_name": "Otoliths",
160
      "display_name": "Otoliths",
            
139
      "id": "d517c6dd-b5fd-4519-92de-cc03e5f30e38",
161
      "id": "d517c6dd-b5fd-4519-92de-cc03e5f30e38",
            
140
      "name": "Otoliths",
162
      "name": "Otoliths",
            
141
      "state": "active",
163
      "state": "active",
            
142
      "vocabulary_id": null
164
      "vocabulary_id": null
            
143
    },
165
    },
            
144
    {
166
    {
            
145
      "display_name": "Regression",
167
      "display_name": "Regression",
            
146
      "id": "de1a22b3-981d-47f5-8c25-d5a05d33fe21",
168
      "id": "de1a22b3-981d-47f5-8c25-d5a05d33fe21",
            
147
      "name": "Regression",
169
      "name": "Regression",
            
148
      "state": "active",
170
      "state": "active",
            
149
      "vocabulary_id": null
171
      "vocabulary_id": null
            
150
    },
172
    },
            
151
    {
173
    {
            
152
      "display_name": "Seals",
174
      "display_name": "Seals",
            
153
      "id": "29aa153e-9f89-42bb-9b39-43ce87f8b0a3",
175
      "id": "29aa153e-9f89-42bb-9b39-43ce87f8b0a3",
            
154
      "name": "Seals",
176
      "name": "Seals",
            
155
      "state": "active",
177
      "state": "active",
            
156
      "vocabulary_id": null
178
      "vocabulary_id": null
            
157
    },
179
    },
            
158
    {
180
    {
            
159
      "display_name": "Wildlife",
181
      "display_name": "Wildlife",
            
160
      "id": "e6b70ffc-8e4c-44fd-a436-02422c77f564",
182
      "id": "e6b70ffc-8e4c-44fd-a436-02422c77f564",
            
161
      "name": "Wildlife",
183
      "name": "Wildlife",
            
162
      "state": "active",
184
      "state": "active",
            
163
      "vocabulary_id": null
185
      "vocabulary_id": null
            
164
    }
186
    }
            
165
  ],
187
  ],
            
166
  "title": "Counting using deep learning regression gives value to 
188
  "title": "Counting using deep learning regression gives value to 
            
167
ecological surveys.",
189
ecological surveys.",
            
168
  "type": "dataset",
190
  "type": "dataset",
            
169
  "url": "https://dataportal.nioz.nl/doi/10.33591/nioz/7b.b.0c",
191
  "url": "https://dataportal.nioz.nl/doi/10.33591/nioz/7b.b.0c",
            
170
  "version": "1"
192
  "version": "1"
            
171
}
193
}