Sea lice mitigation based on historical observations

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for sea lice mitigation. In some implementations, a method includes generating (i) a first record for a first fish indicating an extent of sea lice infestation for the first fish at a first time and (ii) a second record, different than the first record, for a second fish, different than the first fish, indicating an extent of sea lice infestation for the second fish at the first time; and training, based at least in part on the first record and the second record, a model that determines, given one or more input records for a third fish, whether the third fish is likely to be healthy at a second time subsequent to the first time.

What is claimed is: 
     
         1 . A computer-implemented method comprising:
 generating (i) a first record for a first fish indicating an extent of sea lice infestation for the first fish at a first time and (ii) a second record, different than the first record, for a second fish, different than the first fish, indicating an extent of sea lice infestation for the second fish at the first time; and   training, based at least in part on the first record and the second record, a model that determines, given one or more input records for a third fish, whether the third fish is likely to be healthy at a second time subsequent to the first time.   
     
     
         2 . The method of  claim 1 , wherein an extent of sea lice infestation at the first time includes an extent of sea lice for a given fish within a range of time. 
     
     
         3 . The method of  claim 1 , comprising
 determining an extent of sea lice infestation for the third fish based on an image of the third fish;   providing, to the model, a representation of the extent of sea lice infestation for the third fish; and   obtaining, from the model in response to providing the representation, output indicating whether the third fish is likely to be healthy at the second time subsequent to the first time.   
     
     
         4 . The method of  claim 3 , wherein determining the extent of sea lice infestation for the third fish based on the image comprises:
 determining a location of each sea lice on the third fish,   wherein providing, to the model, the representation of the extent of sea lice infestation for the third fish comprises providing an indication of the locations of each sea lice to the model.   
     
     
         5 . The method of  claim 1 , comprising:
 extracting visual features from an image of the third fish;   identifying a record of previous observations of the third fish based on the visual features; and   providing, to the model, both a representation of the previous observations included in the identified record and a representation of an extent of sea lice infestation determined from the image of the third fish.   
     
     
         6 . The method of  claim 1 , comprising:
 selectively initiating sea lice mitigation based on predictions generated by the trained model for the third fish.   
     
     
         7 . The method of  claim 6 , wherein selectively initiating sea lice mitigation based on the predictions generated by the trained model for the third fish includes:
 providing, to a sea lice treatment device, an instruction to treat the third fish for sea lice.   
     
     
         8 . The method of  claim 6 , wherein selectively initiating sea lice mitigation based on the predictions generated by the trained model comprises:
 determining that a predicted health indicator included in the predictions satisfies a mitigation criteria; and   in response to determining that the predicted health indicator satisfies the mitigation criteria, initiating the sea lice mitigation.   
     
     
         9 . The method of  claim 6 , wherein selectively initiating sea lice mitigation based on the predictions generated by the trained model comprises:
 determining that a predicted health indicator included in the predictions does not satisfy a mitigation criteria; and   in response to determining that the predicted health indicator does not satisfy the mitigation criteria, not initiating the sea lice mitigation.   
     
     
         10 . The method of  claim 1 , wherein the first record indicates one or more of an age of the first fish, a weight of the first fish, a size of the first fish, a feature vector of the first fish, or various health metrics of the first fish at the first time; and
 the second record indicates one or more of an age of the second fish, a weight of the second fish, a size of the second fish, a feature vector of the second fish, or various health metrics of the second fish at the first time.   
     
     
         11 . The method of  claim 1 , wherein the first record and the second record indicate conditions of environments in which observations of the first fish and the second fish were made, and
 determining whether the third fish is likely to be healthy at the second time subsequent to the first time is based on a condition of an environment in which an image of the third fish was obtained.   
     
     
         12 . The method of  claim 11 , wherein the conditions of environments in which the observations were made include one or more of location, depth, or temperature. 
     
     
         13 . The method of  claim 1 , wherein (i) the first record indicates whether sea lice mitigation was performed on the first fish and (ii) the second record indicates whether sea lice mitigation was performed on the second fish. 
     
     
         14 . The method of  claim 1 , comprising:
 updating the first record, based on obtaining an image representing the first fish, the first record.   
     
     
         15 . The method of  claim 1 , comprising:
 obtaining an image of the third fish;   providing data representing the image of the third fish to the model; and   obtaining output of the model indicating whether the third fish is likely to be healthy at the second time subsequent to the first time.   
     
     
         16 . A system comprising:
 one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:   generating (i) a first record for a first fish indicating an extent of sea lice infestation for the first fish at a first time and (ii) a second record, different than the first record, for a second fish, different than the first fish, indicating an extent of sea lice infestation for the second fish at the first time; and   training, based at least in part on the first record and the second record, a model that determines, given one or more input records for a third fish, whether the third fish is likely to be healthy at a second time subsequent to the first time.   
     
     
         17 . The system of  claim 16 , wherein an extent of sea lice infestation at the first time includes an extent of sea lice for a given fish within a range of time. 
     
     
         18 . The system of  claim 16 , wherein the operations comprise:
 determining an extent of sea lice infestation for the third fish based on an image of the third fish;   providing, to the model, a representation of the extent of sea lice infestation for the third fish; and   obtaining, from the model in response to providing the representation, output indicating whether the third fish is likely to be healthy at the second time subsequent to the first time.   
     
     
         19 . The system of  claim 18 , wherein determining the extent of sea lice infestation for the third fish based on the image comprises:
 determining a location of each sea lice on the third fish,   wherein providing, to the model, the representation of the extent of sea lice infestation for the third fish comprises providing an indication of the locations of each sea lice to the model.   
     
     
         20 . A computer-readable storage device encoded with a computer program, the program comprising instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
 generating (i) a first record for a first fish indicating an extent of sea lice infestation for the first fish at a first time and (ii) a second record, different than the first record, for a second fish, different than the first fish, indicating an extent of sea lice infestation for the second fish at the first time; and   training, based at least in part on the first record and the second record, a model that determines, given one or more input records for a third fish, whether the third fish is likely to be healthy at a second time subsequent to the first time.